Tourism Economics 2015 21 (1) 9ndash32 doi 105367te20140434

Determinants in tourist expenditurecomposition ndash the role of airline types

BERTA FERRER-ROSELL

Department of Economics Faculty of Tourism University of Girona Edifici Sant DomegravenecPlaccedila Ferrater Mora 1 17071 Girona Spain E-mail bertaferrerudgedu

(Corresponding author)

GERMAgrave COENDERS

Department of Economics Faculty of Economics and Business University of GironaCampus Montilivi 17071 Girona Spain E-mail germacoendersudgedu

ESTHER MARTIacuteNEZ-GARCIA

Department of Economics Faculty of Tourism University of Girona Edifici Sant DomegravenecPlaccedila Ferrater Mora 1 17071 Girona Spain E-mail esthermartinezudgedu

The reduction in transportation costs when travelling with a low-costairline (LCA) seems to have modified the composition of the tripbudget An understanding of expenditure composition whencomparing LCA and legacy airline travellers is vital for destinationmarketers Using micro official statistics data for air travellers toSpain in 2010 and the compositional data analysis (CODA)methodology this study analyses the determinants of trip budgetcomposition and its differences between airline types The authorsconsider transportation expenses as well as basic (accommodationand food) and discretionary (activities shopping etc) at-destinationexpenses Log-ratios of budget share are fitted to a MANOVA withtravellersrsquo attributes as explanatory factors along with the moderatingeffect of the airline type Among the findings are that high-incomeLCA travellers spend relatively more at the destination LCA tourists

The authors are pleased to acknowledge the support of the Spanish Institute of Tourism Studies(Instituto de Estudios Turiacutesticos) in providing them with the raw data Any lack of accuracy or reliabilityin the data analysis is the authorsrsquo responsibility The first author carried out this research with thesupport of a grant from the University of Girona (BR 182011) The authors also wish to thankthe members of the Research Group lsquoStatistics and Data Analysisrsquo and the lsquoResearch Group onCompositional Data Analysis and Other Restricted Sample Spaces CODA-RSSrsquo at the Universityof Girona Juan Luis Nicolau at the University of Alicante and Jorge Gonzaacutelez Chapela at theUniveristy of Girona for their helpful comments on previous versions of this article

TOURISM ECONOMICS10

travelling with friends have a larger share of discretionary expensesand highly educated travellers have a larger share of discretionaryexpenses for both airline types

Keywords tourism demand tourist expenditure low-cost airlinecompositional data analysis (CODA)

Independent tourists represent a growing tourist segment Independent touristsdo not travel on a package deal and often organize the entire trip themselvesThey also have easier access to information and usually prefer to undertake moreactivities In 2004 48 of all tourists to Spain travelled without a packagedeal this rose to 66 in 2010 The increase in the number of independenttourists is accompanied by an increasing expansion and consolidation of no-frillsairlines also referred to as low-cost airlines (LCAs) More LCA users than legacyairline passengers travel independently of package tours In 2010 74 of alltourists who arrived in Spain with an LCA did not book a package deal (IET2004 2010)

The cheaper fares offered by LCAs result in a significant reduction intransportation costs which is expected to modify the composition of the triprsquosbudget (Martiacutenez-Garcia and Raya 2008) This seems to have been the casein Spain a major world tourist destination in 2010 557 of all air travellersarrived by LCA and spent 32 of their trip budget on transportation and 54on accommodation and food In comparison legacy passengers spent relativelymore on transportation (40) and somewhat less on accommodation and food(46) (IET 2010) The reduction in transportation costs can also affect thedistribution of non-transportation expenditures in other words at-destinationexpenditures on accommodation activities and other budget components

LCAs have received growing attention in the literature which either focuseson the demand differences between LCA and legacy airlines (OrsquoConnell andWilliams 2005 Chiou and Chen 2010 Forgas et al 2010 Ferrer-Rosell et al2014) or on LCA demand and its heterogeneity (Castillo-Manzano and Marchena-Goacutemez 2010 Martiacutenez-Garcia and Royo-Vela 2010 Kim and Lee 2011Raya-Vilchez and Martiacutenez-Garcia 2011 Martiacutenez-Garcia et al 2012)However this growing literature on LCA demand and comparisons with legacyairline demand has not to our knowledge focused on tourist expenditure andits composition

Expenditure analyses have been quite frequent in the tourism literature sincethey are of major concern for destination management offices (DMO)marketers tourist agencies and in general to all those engaged in tourismTourist expenditure rather than the number of tourists received is becomingmuch more important for destinations and the economic impact of tourism Theanalysis of expenditure composition provides valuable information fordestination management (over that given by the analysis of absoluteexpenditure) in terms of the type of tourist classified by how they distributetheir travel budget The travel budget can be broken down as with householdbudget studies between non-discretionary and discretionary components Non-discretionary components are accommodation transportation and food wherea minimum amount has to be spent whereas discretionary expenditure includes

11Determinants in tourist expenditure composition

extra activities shopping and so on Depending on personal economic trip andsocio-demographic characteristics tourists may be more or less willing toembark on visits activities excursions shopping and so on and thus theproportion of discretionary tourist budget will change accordingly This distinc-tion between expenditure components is relevant for destinations as they arenaturally more interested in local spending than in expenses paid directly totour operators as is the case for package trips and transportation expensesDMOs are also interested in how the different tourist profiles within LCA andlegacy users allocate their budget For instance if DMOs seek to promoteactivities they should focus their marketing efforts on those tourist typesspending more of their budget on that component and which may differbetween airline types

The objective of this article is to study the determinants of the compositionof tourist expenditure in other words the share of tourism expenditureallocated to the different categories of a trip budget by taking into accounttourist heterogeneity and distinguishing between legacy and LCA marketsegments Two main research questions are addressed

bull How does travelling with an LCA versus a legacy airline affect thedistribution of trip expenses between transportation and other costs and thedistribution of the non-transportation (or also called at-destination) expensesbetween discretionary and non-discretionary

bull Are passenger characteristics affecting budget composition If so do theyaffect it in the same way for both types of airline In other words does theairline type have a moderating effect

With this purpose in mind we build a statistical model explaining budgetcomposition from passenger characteristics along with the moderating effect ofairline type The following budget parts are distinguished

bull proportion or share of total trip expenditure devoted to transportation wheresavings from LCAs arise (amount paid for transportation from the airportof origin to the point of accommodation and the return trip)

bull share of total trip expenditure devoted to accommodation and food (basicthat is non-discretionary)

bull share of total trip expenditure devoted to doing activities moving aroundat the destination and shopping (discretionary tourist expenditure)

Statistical analysis of budget compositions is a methodologically challengingtask Shares in budgets as any other composition are expressed as proportionsor percentages of a total whose sum can only be 1 or 100 Compositional datalie in a restricted space and only convey information regarding the relative sizeof components to one another Aitchisonrsquos (1986) seminal work started a fruitfultradition in compositional data analysis (CODA) and of which the most widelyused technique is the transformation of compositions This is achieved by meansof logarithms of ratios Working with log-ratios has not only methodologicalimplications but also substantive ones Without log-ratios components areestimated and interpreted separately from one another as if they could varyindependently (ceteris paribus) which is impossible the relative importance of

TOURISM ECONOMICS12

one component (budget share) can increase only if the relative importance ofat least one other decreases Most methods used to model budget share suchas the almost ideal demand system (Deaton and Muellbauer 1980) ignore atleast partly the constraints and distributional nature of compositional data Tothe best of our knowledge there is no scholarly study of the composition oftourist budget using an appropriate methodology for compositional data analysis

This article is structured as follows First we present a review of the mainapproaches in modelling tourist expenditure which is then followed by adescription of the CODA methodology Then we introduce the method anddata which are followed by the results and finally by the overall conclusionsand discussion

Major approaches in modelling tourism expenditure

The study of expenditure in terms of composition (relative share of each partof the budget) is not the same as the study of expenditure in absolute termsand the variables affecting absolute expenditure may differ from those affectingrelative expenditure Tourism budgets have been approached from bothperspectives in the literature More precisely tourism expenditure has beenanalysed globally in absolute terms per budget parts in relative terms (share)per budget parts and as a part in itself in family budgets

The vast majority of microeconomic tourism demand studies (24 out of 27in the review of Wang and Davidson (2010)) concern the prediction of onesingle aggregated expenditure variable Methods range from mean comparisontests (Craggs and Schofield 2009) OLS and WLS regression (Downward andLumsdon 2000 2003 Cannon and Ford 2002) to advanced econometrictechniques For instance Hung et al (2012) use quantile regression to buildequations predicting not only typical expenditures but also the highest andlowest Alegre et al (2009) Eugenio-Martin (2003) Hong et al (1999) andNicolau and Maacutes (2005) propose double-hurdle Heckit and related models toseparate the decision whether to spend on tourism from the decision of howmuch to spend With respect to the explanatory variables used the mostcommon are income age gender marital status education place of residencelength of stay travel group size and composition accommodation main trippurpose and activities (Marcussen 2011) In general explanatory variables canbe grouped into economic variables (prices and income) socio-demographicvariables and trip or travel-related variables (Wang et al 2006 Sainaghi 2012)Nicolau (2009) includes individual price sensitivity estimated in a previousmodel

Another stream of research is that which analyses tourist expenditure pertourism product (for example lodging food transportation and sightseeingentertainment) A common argument for studying tourist expenditure patternsper tourism product is that it provides vital information to travel organizersand destination marketers when designing the appropriate marketing strategiesFor this purpose researchers have used several methods such as Tobit MANOVAor seemingly unrelated linear regressions (Pyo et al 1991 Cai et al 1995Oppermann 1996 Cai 1998 1999 Lee 2001 Lehto et al 2001 Jang et al2004 Wang et al 2006) Socio-demographic variables are the most common

13Determinants in tourist expenditure composition

significant variables Age affects expenditure on meals as does travel groupwhich also affects transportation expenditure Marital status usually affects foodand accommodation expenditure

Since part expenditure in absolute terms is related to total expenditure acommon finding in these studies is that some of the explanatory variables affectall budget elements roughly equally For example Cai et al (1995) found thatthe higher the level of education the higher the expenditure in all budget partsand Wang et al (2006) conclude similarly about household income Such resultswill never be obtained when analysing budget share

The empirical analysis of budget share both for tourism expenditure andgeneral family budgets commonly involves estimating an almost ideal demandsystem of equations (Deaton and Muellbauer 1980) The almost ideal demandsystem is designed to analyse the interdependences of budget allocations thusovercoming the limitations of single-equation modelling and has receivedmuch attention in the last decade (Song et al 2012) It directly fits shares(composition) as dependent variables in a set of simultaneous regressions Theapproach has focused mainly on estimating price and income elasticities Studiesthat use macroeconomic data include among others those by Divisekera (20072009 2010) Fujii et al (1985) OrsquoHagan and Harrison (1984) Syriopoulos andSinclair (1993) and Wu et al (2011)

The almost ideal demand system has been applied both to data from a givenorigin to multiple destinations (that is ex ante ndash see Li et al 2004 Divisekera2009) or from multiple origins to a given destination which is the case in ourarticle (ex post ndash see Divisekera 2009) In ex post studies it is commonly assumedthat various commodities can be aggregated to broad bundles of productsprovided that prices in a bundle move in parallel and that the utility functionwith respect to tourism and other goods is weakly separable This makes itpossible to conceptualize tourism consumption as a multi-stage process In thefirst stage tourists allocate a household budget part to tourism consumptionin the second a tourism budget part to each tripdestination in the third adestination budget part to each good and service including transportationaccommodation and so on Ex post studies such as ours model only the laststage There are however studies of how tourism competes against othercategories of discretionary expenditure using individual micro data Forexample Melenberg and Van Soest (1996) use different parametric and semi-parametric Tobit models to explain the vacation budget share from householdcharacteristics and Dolnicar et al (2008) analyse how households allocatediscretionary income between tourism and competing uses

The review of Wang and Davidson (2010) concludes that since the vastmajority of tourism demand studies are conducted at macro level (this holdseven more for budget share analysis) there is room for more micro-econometricstudies in this area as the only manner of accounting for demand heterogeneityThe almost ideal demand system approach when applied to micro data caninclude individual characteristics For example Coenen and van Eekeren (2003)and Fleischer et al (2011) make ex ante studies of individual budget share inSweden and Israel respectively and use previous Heckit-type selectionequations to model the decision whether to travel or not Coenen and vanEekeren (2003) include household size and income as individual characteristicsFleischer et al (2011) add age education real state ownership place of birth

TOURISM ECONOMICS14

and Internet use The basic aim of those articles is to estimate elasticities andthey use the individual characteristics only as controls Fleischer et al (2011)are the only researchers to provide the estimates of equations predicting budgetshare from traveller characteristics as a by-product of their elasticity estimatesBeing born in the country of origin is reported to increase the transportationshare and reduce the share of on-site expenditures Education and householdreal estate assets reduce the share of on-site expenditures

The aim and approach of this article are similar to those of Coenen and vanEekeren (2003) and of Fleischer et al (2011) regarding the use of individualcharacteristics to predict budget share but differ in three important respects

bull Our study is ex post so that Heckit modelling is unfeasiblebull The effect of individual characteristics on budget composition (share) is the

core of the analysis Explanatory variables are individual characteristics ratherthan prices The results will include the effects of traveller heterogeneityrather than demand equations

bull Our analysis takes into account the compositional restrictions of the databy using the CODA methodology

The CODA methodology

Compared to absolute data compositional data such as budget share lie in aconstrained space A D-term composition measured on individual i xi1 xi2hellipxiD

has the following constraints

0 le xid le 1 and DΣ

d=1xid = 1 (1)

Aitchison (1986) and Pawlowsky-Glahn and Buccianti (2011) warn against theserious problems that arise when using standard statistical analysis tools oncompositional data Compositional data are non-normal and heteroscedasticOne component can increase only if some other(s) decreases This results innegative spurious correlations among the components and prevents interpretingeffects of linear models in the usual way lsquokeeping everything else constantrsquo

Even if specialized CODA techniques are starting to appear (for exampleRonning 1992 Thioacute-Henestrosa and Martiacuten-Fernaacutendez 2005) the easy way(Aitchison 1986 McLaren et al 1995 Fry et al 1996) involves transformingcompositional data so that they can be subject to standard and well-understoodstatistical techniques This is the approach we take in this article In short thisinvolves using the transformed share by means of logarithms of ratios insteadof the raw share

To ensure compositional coherence of predicted budget share (unit sum andnon-negativity) a set of parameter constraints is imposed to the almost idealdemand system However the presence of an error term with an unboundeddistribution (usually normal) results in a non-zero probability that actual sharelies outside the [01] interval (McLaren et al 1995 Fry et al 1996 Fry 2011)In other words the in fact bounded distribution of budget share results in amisspecification of the almost ideal demand system and of any model fittingpercentage share with an unbounded error distribution

15Determinants in tourist expenditure composition

The fact that the error term in any proper system of demand equationsapplied to budget share should take into account the data compositional naturehas been widely acknowledged (Aitchison 1986 Ronning 1992 Fry et al1996) However the review in Fry (2011) reports few studies that have appliedthe CODA methodology to demand equations and to the study of householdbudgets (McLaren et al 1995 Fry et al 1996 2000)

When fitting demand equations to log-ratios of expenditure componentsEngelrsquos aggregation condition which requires that household total expenditureshould equal the sum of components is automatically satisfied The model caneasily be extended to a full consumer demand analysis by the introduction ofprices and other covariates such as consumer characteristics (Aitchison 1986)McLaren et al (1995) relate the CODA analysis with log-ratio transformationto the almost ideal demand system and conclude that CODA makes it possibleto reach the same objectives with normal and homoscedastic error terms

Related developments are the indirect addilog system (Houthakker 1960)and the generalized addilog system (Bewley 1982) When applied tocompositions (for example Bewley and Fiebig 1988) the latter is equivalentto the CODA methodology in which the log-ratios of each component over thegeometric means of all components are the dependent variables in the set ofsimultaneous regressions

To the best of our knowledge there is no study of tourism budget share usingthe CODA methodology or any other methodology accounting for thecompositional constraints in compositional data

Method and data

Statistical approach

The simplest CODA approach involves applying standard statistical techniqueson logarithms of ratios of components Several log-ratio transformations havebeen suggested in the early CODA literature (Egozcue et al 2003) The additivelog-ratio transformation (alr) used by Fry et al (1996 2000) is the most popularand the easiest to compute given that it is simply the log-ratio of eachcomponent to the last

yid = ln(xidxiD) = ln(xid) ndash ln(xiD) with d = 123D ndash 1 (2)

The centred log-ratio transformation (clr) used by the generalized addilogdemand system (Bewley 1982) computes the log-ratios of each component overthe geometric mean of all the components including itself

⎛ xid ⎞yid = ln ⎜ ndashndashndashndashndashndashndashndashndashndashndashndash ⎟ with d = 123D (3)

⎝ D xi1xi2xi3 xiD ⎠

All log-ratio transformed yid variables recover the full unconstrained ndashinfin to infinrange It must be noted that one dimension is lost in the alr while in the clrone dimension is a linear combination of the remaining

TOURISM ECONOMICS16

The alr is commonly used for statistical modelling and prediction ofcompositions (Fry et al 1996) Conversely the clr transformation is commonlyused for statistical techniques which are based on a metric such as a clusteranalysis because of its preservation of distances even though it leads to asingular covariance matrix Thus while the alr would be appropriate for thepurpose of this article the fact that one component must be used as referencefor all others reduces its flexibility and interpretability Alternatives arepresented at the end of this subsection

While having zero expenditures in absolute data indeed has significantmethodological consequences (Lee 2001) these consequences are arguably moreserious in the CODA methodology If the xid variables contain zeros then log-ratios cannot be computed An obvious initial procedure to reduce zeros is toamalgamate small and conceptually similar components with many zeros intolarger ones In tourism budget research it can be useful to group together allexpenditure on activities or all expenditure on food for instance

In certain instances some zero components result from individual character-istics which are called essential zeros in the CODA literature (Aitchison 1986)Another typology of zeros encountered in the CODA literature is the roundingzero that is a component which is present but is too small to be detected bythe measurement instrument This is typical in chemical biological andgeological compositions and the CODA literature offers ample instruments todeal with rounding zeros

A classic essential zero example in economics is in household budget researchwhen measuring expenditure on tobacco and will essentially be zero if allmembers are non-smokers If tobacco expenditure is decidedly in the researchersrsquointerest the target population should be redefined to include only smokers Inmany instances budget research is often not clear whether zero expenditurescome closer to being essential or rounding zeros In some cases they can beunderstood as corner solutions in a utility maximization problem In others theycan be understood as the inherent randomness of human behaviour or as thelimitations of the data Tourists may spend a certain amount on activities oncertain trips but not on others and so surveys of only one trip will unavoidablycontain some zeros of this type Tourists may also forget or fail to report trivialexpenses such as postcard shopping local bus tickets and going to a museum(see Legoheacuterel (1998) for a discussion on the instability of tourism expenditure)Fry et al (2000) claim that in both situations zeros can be proxied by a verysmall value and thus be treated as rounding zeros In tourist budget researchto treat zero expenses in activities as rounding zeros implies assuming firstthat there is basically no tourist type who will never spend anything onactivities and second that tourists who generally spend little on activities arebasically similar to those who spend nothing or fail to report small expensesWe find both assumptions to be reasonable

Fry et al (2000) essentially used the same zero replacement strategy that waslater suggested by Martiacuten-Fernaacutendez et al (2003) namely replacing xid = 0 with

xprimeid = kδid with 0 lt k lt 1 (4)

where δid is the smallest detectable proportion for individual i and component d

17Determinants in tourist expenditure composition

Martiacuten-Fernaacutendez et al (2003) suggest using k = 065 although a sensitivityanalysis of the results on the choice of k is always advisable (we used k = 030and k = 099 with no sizeable change in the estimates) Next non-zero xid valueshave to be reduced in order to preserve the unit sum and the ratios among non-zero components As suggested by Martiacuten-Fernaacutendez et al (2003) with

xprimeid = xid (1 ndash Σxid=0

xprimeid) (5)

Simulations show this method performs particularly well if the proportion ofzeros is below 10 (Martiacuten-Fernaacutendez et al 2011)

In our data set zeros were present only in one budget category (discretionaryexpenditure) The minimum amount spent by the non-zero group was euro1which roughly corresponds to the price of a city bus ticket the entrance to asubsidized local museum or a cheap souvenir Since the total expenditureis known for each individual we compute δid by dividing euro1 with the totalexpenditure of individual I (see the appendix for the SPSS commandsyntax)

In this article we consider alternative log-ratio transformations which aremore flexible than the alr and clr in that the denominator does not have tobe the same in all ratios This increased flexibility makes it easier to computelog-ratios which are more interpretable with respect to the researchersrsquo questions orhypotheses

In general an interpretable log-ratio transformation is easy to computewhenever there is an interpretable sequential binary partition of componentsinto pairs of groups of components according to the researchersrsquo objectives orto the conceptual similarity of the components These partitions start bydividing components into two clusters and then continue by subdividingone of the clusters into two until each component constitutes its owncluster D components always involve Dndash1 partitions These partitions are bestunderstood as a partition tree or dendrogram (Pawlowsky-Glahn and Egozcue2011)

A meaningful log-ratio transformation takes ratios of the geometric meansof the two component clusters at each partition Numerators and denominatorsare interchangeable In our article we consider xi1 = transportation expenditurexi2 = accommodation and food (basic expenditure) and xi3 = activities andshopping (discretionary expenditure) An interpretable sequential partition isshown in Figure 1

The sequential partition in Figure 1 can have a dual interpretation It mayimply that researchers consider both types of at-destination expenses to bemutually similar and less similar to transportation Or it may imply that theresearch questions involve the distribution of total expenditure betweentransportation and at-destination expenditure and the distribution ofat-destination expenditure into basic and discretionary components as is thecase in our article

The first log-ratio compares transportation expenditure with the geometricmean of accommodation and food (basic expenditure) and activities andshopping (discretionary expenditure) With this ratio we want to observe howtravelling with an LCA or a legacy company affects the share of transportationcompared to non-transportation (at-destination) expenses

TOURISM ECONOMICS18

Figure 1 Sequential partition of total expenditure

⎛ xi1 ⎞ 1 1yi1 = ln ⎜ ndashndashndashndash ⎟ = ln(xi1) ndash ndash ln(xi2) ndash ndash ln(xi3) (6)

⎝ xi2xi3 ⎠ 2 2

Positive values show a transportation share greater than the geometric meanof the remaining two components Negative values show the opposite

The second log-ratio is a ratio of accommodation and food (basic expenditure)over activities and shopping (discretionary expenditure) With this ratio wewant to find out to what tourists allocate the rest of their trip budget andwhether more is allocated to basic or discretionary expenditure once they havepaid for (often in advance) their transportation

⎛ xi2 ⎞yi2 = ln ⎜ ndashndashndash ⎟ = ln(xi2) ndash ln(xi3) (7)

⎝ xi3 ⎠

Positive values show a basic (accommodation and food) share which is largerthan the discretionary (activities and shopping) share Negative values illustratethe opposite

Log-ratio transformations based on sequential binary partitions areproportional to the isometric log-ratio transformation (ilr see Egozcue et al2003 Egozcue and Pawlowsky-Glahn 2005)

As compositions are vector variables they cannot be analysed component-wise by means of univariate regression ANOVA models and the likeSeemingly unrelated regression models for continuous explanatory variables ormultivariate ANOVA (MANOVA) models for categorical explanatory variablesare appropriate MANOVArsquos multivariate tests and statistics (for example

x2 x3 Transportation

Global partition of total expenditure

x1

Accommodation

amp food

(basic)

Activities

amp shopping

(discretionary)

Partition of

expenses other

than transportation

(at-destination

expenses)

19Determinants in tourist expenditure composition

Pillairsquos trace Hotellingrsquos trace or Wilkrsquos lambda) are invariant to howcomponents are arranged in Figure 1 to compute log-ratios (Mateu-Figueras etal 2013) Univariate tests referring to each particular log-ratio are not invari-ant hence the importance of the interpretability of each log-ratio The softwareSPSS 19 is used to estimate the MANOVA model (GLM procedure)

We have to distinguish between three types of variables moderatingexogenous and endogenous to be included in the analysis

Moderating variables are those which modify the effect of exogenousvariables In our article we treat type of airline as a moderator in other wordswe include its interaction terms with all other variables in the MANOVA model

Exogenous variables are assumed to affect expenditure but not the converseSocio-demographic characteristics (travellerrsquos age gender education incomeetc) are obviously determined prior to any travel decision and hence exogenouswith respect to expenditure

Many of the choices tourists make are interdependent (Dellaert et al 1997)or at least planned simultaneously (Fesenmaier and Jeng 2000) So expenditureis arguably decided on at the same time as many trip attributes At least itis unclear whether expenditure is consistently decided on after trip attributesby all travellers We consider them to be endogenous variables and thereforedo not include them in the MANOVA model as explanatory in order to preventendogeneity problems Instead we display them graphically in the log-ratiospace The means of both log-ratios within each endogenous category are thecategory coordinates

The large sample size (see next subsection) makes it possible to use lowp-values Moderating effects with p-values higher than 001 according to eitherPillairsquos trace Hotellingrsquos trace or Wilksrsquo lambda were removed from the modelAll the variables and moderating effects included in the final model aresignificant at 001

Sample and variables

In this article we use secondary official statistics data The data were providedby the Instituto de Estudios Turiacutesticos (IET) an official agency of the Ministryof Industry Energy and Tourism and one which produces the majority oftourism data in Spain The survey is known as the Encuesta de Gasto Turiacutestico(EGATUR) in which tourism expenditure and other tourist information suchas trip information and tourist sociodemographic characteristics are studiedThe EGATUR survey conducted in 27 major Spanish airports in 2010 usedCAPI (computer assisted personal interview) to interview tourists leaving thecountry The sample is non-proportionally stratified by country of residenceairport and month See IET (2012) for further details on the EGATURmethodology

Our universe is a subset of the EGATUR universe which consists ofEuropean leisure visitors arriving by air and spending at least one night inSpain We excluded flights from outside Europe because LCAs mostly operateshort-haul flights We also centred our study on only those trips with one singledestination thus excluding multi-stage trips as the decision process regardingexpenditure composition for these trips is expected to fundamentally differ fromthat of single-stage trips Stays of over 120 days were also excluded

TOURISM ECONOMICS20

For this study we did not consider tourists

bull who have essential zeros in accommodation (tourists who own a house at thedestination or tourists who stay with friends or relatives)

bull who do not decide how much they will spend on certain components(business and study trips)

bull who do not pay for the trip themselves (trips paid for by employers familyfriends prizes offers etc)

bull for whom composition is wholly or partly unobserved (package tourists)

The final sample size was N= 19359From the expenditure variables included in the EGATUR survey database

and used in the model as budget components we first put the amount paidfor transportation (x1) This component has no zeros

Second the amount of money paid for accommodation and food isundistinguishable for full-board half-board or bed and breakfast accommodationTherefore we define a joint accommodation and food component (basic expendi-ture) In this component we include not only the amount paid for consumptionin bars and restaurants but also for buying groceries and everyday products insupermarkets (x2) This component has no zeros

Finally EGATUR provides an aggregated expenditure on activities andshopping (except groceries and everyday products) To this we added theconceptually similar amount paid for moving around the destination (publictransportation andor car rented at the destination) in order to build anactivities and shopping component (x3) This component had 98 zeros whichstresses its discretionary character Zeros were replaced as explained in thestatistical approach subsection

We consider as at-destination (or also called non-transportation) expenses thebasic component (food and accommodation) and the discretionary component(activities and shopping)

Share and log-ratios are described in Table 1 Explanatory variables are thelevel of education income country of residence gender travel groupprofessional status (for pensioners we consider their last professional position)and age The age variable is built as a combination of the original age variableand the variable referring to the individual economic situation This is donein order to have a specific pensioner category taking into account the varyingretirement ages across countries and professions and thus capturing thosetourists who have more free time to undertake a trip regardless of physical ageTable 2 shows their categories and frequencies

As endogenous variables which are not included in the MANOVA modelbut are included in the log-ratio space we include activities undertaken at thedestination accommodation length of stay total expenditure quartilesquartiles of expenditure made at destination per day (basic plus discretionaryexpenses) and a combination of motivation and destination This last variablehas been constructed as a combination of the motivation and the destinationvariable We distinguish between first those tourists travelling for culturaltourism to singular cities second those who come just for leisure either inthe countryside or more commonly at the seaside and third other typologies(Table 3)

21Determinants in tourist expenditure composition

Table 1 Percentage share and log-ratio descriptive statistics

Min Max Mean SD Skewness Kurtosis

Transportation component (x1) 003 9611 2846 1387 0837 0789Basic component (x2) 021 9963 5472 1613 ndash0114 ndash0450Discretionary component (x3) 001 8668 1681 1292 0988 1174Transportationat-destinationlog-ratio (y1) ndash757 548 01506 10702 0714 1404

Basicdiscretionary log-ratio (y2) ndash530 872 17657 18173 1573 2151

Results

Results for the exogenous variables

Table 4 presents the MANOVA results The univariate R2 corrected for thedegrees of freedom is 0045 (first log-ratio) and 0105 (second log-ratio) Themultivariate uncorrected R2 is 1 ndash Λ = 0178 and corrected for the degrees offreedom is 0176 We checked that the results did not change when removingthe four outliers with the highest Cookrsquos distance

Table 4 presents the estimates for the two log-ratios the log-ratio oftransportation over the other two categories (y1) and the log-ratio of basicexpenditure (accommodation and food) over discretionary expenditure (activitiesand shopping) (y2) respectively Because of the moderating effects within eachlog-ratio there are two columns representing the effects when flying with anLCA or with a legacy airline (main and moderating effects have already addedtogether for easier reading) If the LCA and legacy columns are different thereis a significant moderating effect (p-value lt 001)

As we are interpreting log-ratios a positive estimate of a given predictorcategory means that tourists in that category spend more on the numeratorcompared to the denominator than tourists in the reference predictor categoryA negative estimate shows the opposite

The intercept term shows the main effect of company type that is thepredicted log-ratio for each type of airline within the reference category for allvariables (university education medium income resident in the UK or Ireland25ndash44 yearsrsquo old male travelling with partner and mid-level employee) Withinthe reference categories legacy users spend a higher proportion on transportationcompared to other expenses and LCA users spend more on basic expensescompared to discretionary

The level of education seems to have more to do with activities undertakenthan with transportation Results show that a lower level of education resultsin higher expenditure on basic expenses compared to discretionary and is almostequal for users of both types of airline For mainly or only LCA users a lowerlevel of education results in a higher share in transport compared to the othertwo categories If we put it in another way there seems to be a distinct highlyeducated LCA user segment which utilizes the savings in transport toincrease expenditure in non-transportation expenses and even more so indiscretionary

TOURISM ECONOMICS22

Table 2 Frequency distributions of exogenous and moderating variables in the MANOVAmodel

Variables Count Percentage

Level of educationUp to high school 6019 311Universitya 13340 689

Income categoryMediumlow 797 41Mediuma 13186 681Mediumhigh 4376 226High 1000 52

Country of residencePortugal 624 32Other European countries 850 44Belgium 1006 52Netherlands 1069 55Austria Switzerland and Liechtenstein 1122 58France 1326 68Scandinavian countries 1789 92Germany 2363 122Italy 3113 181United Kingdom and Irelanda 6097 315

GenderFemale 8530 441Malea 10829 559

Travel groupAlone 2484 128In family 2980 154With friends 4440 229With partnera 9446 488

Professional statusHomemaker 517 27Unemployed 602 31Student 1432 74Low-level employee 915 47Mid-level employeea 10964 586High-level employee 2138 110Self-employed 2791 144

AgeOver 45 yearsrsquo old and pensioner 1484 77Over 45 yearsrsquo old and not pensioner 5009 25925ndash44 yearsrsquo olda 10569 54615ndash24 yearsrsquo old 2297 119

Type of airline (moderating variable)Legacy 5884 304Low cost 13475 696

Note aReference categories in the MANOVA model chosen either because they are the largestcategories or because they are the most standard conceptually considered

23Determinants in tourist expenditure composition

Table 3 Frequency distributions of endogenous variables (not included in the MANOVAmodel)

Variables Count Percentage

ActivitiesGolfYes 227 12No 19132 988HikingYes 364 19No 18995 981Sporting eventsYes 603 31No 18756 969Nautical sportsYes 1645 85No 17714 915Other sportsYes 904 47No 18455 953Cultural visitsYes 11180 578No 8179 422Cultural eventsYes 2452 127No 16907 873Other cultural activitiesYes 4175 216No 15184 784SpaYes 1199 62No 18160 938Theme parksYes 1745 90No 17614 910GastronomyYes 1565 81No 17794 919Type of accommodationHotel 4ndash5 5056 261Hotel 3 4922 254Hotel lt 3 6673 345Rented appartment 2116 109Other accommodation 592 31Length of stay5 or less nights 9911 5126ndash8 nights 5952 3079ndash12 nights 1592 8213ndash15 nights 1261 6516 and more nights 643 33

TOURISM ECONOMICS24

Table 3 continued

Variables Count Percentage

Total expenditureQuartile 1 5074 262Quartile 2 5104 264Quartile 3 4986 258Quartile 4 4195 217Expenditure at destination per dayQuartile 1 4855 251Quartile 2 4832 250Quartile 3 4868 251Quartile 4 4804 248Motivation-destinationLeisure on coastin countryside 8083 418Urban tourism 5324 275Others 5952 307

Level of income in contrast affects both log-ratios The higher the incomethe lower the transport share The higher the income the lower the basicdiscretionary expense ratio that is a higher portion of the non-transportexpense is devoted to the discretionary budget

As far as the country of residence is concerned the results show that it affectsmainly the basicdiscretionary log-ratio and exhibits strong airline typemoderating effects LCA users tend to differ depending on their country ofresidence when distributing non-transport expenses into basic versusdiscretionary In general when compared to the UK and Ireland residentcategory (reference category) all other European residents have a largerdiscretionary share More precisely the French Italians and Portuguese have thehighest discretionary share within LCA users and other European countries(mainly central and eastern European) have the highest discretionary sharewithin legacy airline users

Gender has a small effect which is constant for both types of airline Femaletravellers tend to slightly increase the transport share and the share of basicexpenses within the non-transport expenses

The travel group seems to affect both log-ratios and has a moderating effectwith airline type Those tourists who travel in a family group have a highershare in transport expenses when flying with legacy airlines In the second log-ratio travelling as a family with legacy airlines increases the basic expensesshare compared to the share in discretionary Travelling with friends increasesthe share in non-transport expenses for both airline types Besides those whotravel with friends spend more in the discretionary share compared to basicexpenses and those who fly with LCA even more so Finally travelling aloneincreases the basic expenses share compared to the share in discretionaryexpenses for both airline types

When professional status is considered the results show that it mostly affectsthe basicdiscretionary log-ratio and there are some relevant airline type

25Determinants in tourist expenditure composition

Table 4 MANOVA results

y1 log-ratio y2 log-ratiotransportation basicat-destination discretionary

LCA Legacy LCA Legacy

Intercept 0225 0274 2195 1881

Up to high school 0192 0078 0406 0437University 0 0 0 0

Mediumlow income 0073 0289 0037 0518Medium income 0 0 0 0Mediumhigh income ndash0227 ndash0058 ndash0269 ndash0006High income ndash0495 ndash0190 ndash0463 ndash0141

Residence in Portugal 0041 0069 ndash1136 ndash0580Residence in other European Countries 0326 ndash0012 ndash0772 ndash0820Residence in Belgium ndash0127 ndash0159 ndash0667 ndash0461Residence in Netherlands ndash0120 ndash0197 ndash0678 ndash0530Residence in Austria + Switzerland +Liechtenstein ndash0017 ndash0087 ndash0687 ndash0436

Residence in France ndash0112 ndash0111 ndash0870 ndash0553Residence in Scandinavian countries ndash0045 ndash0038 ndash0660 ndash0477Residence in Germany ndash0283 ndash0176 ndash0452 ndash0392Residence in Italy ndash0209 ndash0266 ndash0855 ndash0503Residence in UK or Ireland 0 0 0 0

Female 0094 0094 0093 0093Male 0 0 0 0

Travelling alone ndash0053 ndash0002 0081 0126Travelling in family 0085 0193 0014 0213Travelling with friends ndash0237 ndash0191 ndash0504 ndash0282Travelling with partner 0 0 0 0

Homemaker 0080 0160 0253 0235Unemployed 0030 0159 ndash0036 ndash0039Student ndash0005 ndash0104 ndash0331 ndash0624Low-level employee ndash0034 0194 ndash0194 0063Mid-level employee 0 0 0 0High-level employee ndash0008 ndash0064 ndash0001 ndash0139Self-employed 0029 ndash0086 0193 ndash0072

Over 45 yearsrsquo old and pensioner ndash0060 ndash0060 0633 0633Over 45 yearsrsquo old and not pensioner ndash0010 ndash0010 0142 014225ndash44 yearsrsquo old 0 0 0 015ndash24 yearsrsquo old 0002 0002 ndash0019 ndash0019

moderating effects Legacy users with a low professional status (homemakerunemployed or low-level employee) spend more on transport compared to thenon-transport expenses Conversely students using legacy airlines spend moreon non-transport expenses and these non-transport expenses are to a greater

TOURISM ECONOMICS26

Figure 2 Endogenous variables on the log-ratio space Mean log-ratio valueswithin each category of the endogenous variables

extent devoted to discretionary expenses compared to basic expensesHomemakers increase basic expenditure compared to discretionary with respectto the reference category (mid-level employee) for both airline types LCA userswho are self-employed also increase basic expenditure compared to discretionaryLow-level employees flying with an LCA increase the discretionary sharecompared to the basic and high-level employees flying with legacy airlines alsospend relatively more on the discretionary part

The results show that age affects the second log-ratio more than the firstand there is no moderating effect with the airline type Being a pensioner resultsin a higher share in basic expenses compared to discretionary expenses for bothairline types The remaining categories have hardly any differences from thereference 25ndash44 yearsrsquo old

Results for the endogenous variables

As introduced in the methodology section Figure 2 shows how the endogenousvariables behave in the log-ratio space The origin of the graph is representedby the mean log-ratios in Table 1 01506 for the transportother ratio and17657 for the basicdiscretionary ratio and according to this the graph showscategories with higher or lower ratios than the average

075

060

045

030

015

000

ndash015

ndash030

ndash045

ndash060

Ratio tra

nsport

ationa

t-destination e

xpenses

Ratio basicdiscretionary expenses

050 075 100 125 150 175 200 225 250

27Determinants in tourist expenditure composition

Regarding accommodation it seems that rented apartments and other typesof accommodation behave similarly and have the lowest budget share intransport In the case of hotels as the star rating increases expenditure on basiccompared to discretionary increases and expenditure on transport compared toat-destination decreases

As far as the length of stay is concerned longer stays increase the share ofnon-transport expenditure in basic expenses and reduce the share in transport

As regards destination and motivation tourists going to a seaside or country-side destination for leisure purposes behave quite differently from those goingto an urban destination for urban or cultural tourism The former group spendsmore on transport and less on the discretionary portion However those comingfor urban tourism show an increase in the share of non-transport expenses aswell as in the discretionary share compared to basic expenses

The total expenditure variable gives relevant information It is ordered alongthe first log-ratio axis from top to bottom meaning that those tourists havinga lower total expenditure spend relatively more on the transport share and thosewho have a higher total expenditure spend more on the other two parts of thebudget Regarding the expenditure made at destination per day it is ordereddiagonally from the upper-right corner to the bottom-left corner meaning thattourists with a higher at-destination expenditure per day spend less ontransportation and at the same time more on the discretionary part within at-destination expenses

Finally activities undertaken are located in the lower left quadrant of thegraph meaning that undertaking activities (discretionary expenses) decreasesthe share in transport expenses and although they continue to spend more onbasic expenses than on discretionary the discretionary share increases We havehighlighted some of them Those tourists attending sports and cultural eventsare those spending the most on discretionary compared to basic expensesHiking or some nautical sports could be free of charge thus tourists doing theseactivities have a basic to discretionary ratio close to the mean As for thetransportation share all activities tend to reduce this share about equally withthe exception of golf which leads to a much more substantial reduction

Conclusion and discussion

The main purpose of this article was to study the composition of touristexpenditure and its determinants that is the drivers of the share of touristexpenditure allocated to the different categories of a travel budget and withspecial emphasis on the distinction between legacy versus LCA travellers

The main differences between legacy and LCA users are that income somecountries of residence the travel group and some occupations have differenteffects depending on the type of airline Tourists with a medium or low incomewho travel with legacy airlines tend to spend a greater share of their travelbudget on transportation and basic expenses but those travelling with LCAstend to spend more at the destination and more specifically ondiscretionary expenditure LCA tourists residing in the lsquoother Europeancountriesrsquo spend more on transportation Families flying with legacy airlinesspend more on the transportation share compared to the at-destination share

TOURISM ECONOMICS28

and also spend more on basic expenses compared to discretionary ones Finallyin terms of occupational status legacy users with low-level employment spendmore on transportation whereas the opposite occurs with LCA travellers whospend more on the discretionary component Students and high-level employeesflying with legacy airlines spend more on the discretionary part of the tripbudget

As far as implications for management are concerned assuming that theobjectives of DMOs are to increase the allocation to the at-destinationcomponent of the trip budget (that is lowering the first log-ratio) some of therecommended actions would be to increase marketing efforts directed atmedium- and high-income earners especially at those travelling on LCAs toincrease marketing strategies in some of the European outbound markets suchas Germany (especially those using LCA) and Italy and the Netherlands(especially those flying with legacy airlines) DMOs should also focus on thosetourists travelling with friends (airline type is inconsequential) and studentsusing legacy companies As for the endogenous variables DMOs should directtheir efforts towards those tourists undertaking activities towards those stayinga little bit longer than a week and towards those with high total expenditureFurthermore if DMOs are also interested in capturing the groups who spendrelatively more on discretionary expenses (lowering the second log-ratio) theyshould focus their marketing efforts on those tourists who are highly educatedas well as those who have medium to high incomes and who use LCAsAdditionally they should focus on markets other than the UK and IrelandYoung tourists and tourists who travel with friends (where airline type isinconsequential) low-level employees and students using LCAs as well asstudents and high-level employees flying on legacy airlines should also betargeted In relation to the endogenous variables DMOs must take into accountthose tourists undertaking payable activities those on shorter stays or comingfor urban and cultural tourism or other purposes

The appeal of the CODA methodology for studying tourism budgets lies inthe fact that once the variables have been transformed the researcher can usestandard and well understood statistical models with unbounded error termdistributions while ensuring that the results will be compositionally coherentand that the standard statistical assumptions will hold Conversely results arenot compositionally coherent when we use the MANOVA model on raw sharewithout transforming it into log-ratios We have seen with our data that 84of individuals included in the sample have at least one limit of the 90 shareprediction intervals outside the [01] range

Contrary to the classic and highly restrictive alr and clr transformations theCODA methodology offers the potential to construct tailor-made log-ratios whichare intuitive to interpret and suit the research questions at hand A classificationtree of components is a clear and useful tool in this respect We encourageresearchers to use this approach in further research on tourism expenditure whenbudget share division is fundamental to the researchersrsquo questions

In this study we have encountered two limitations First contrary to studiesthat use macro data in micro data studies the coexistence of full-board halfboard bed and breakfast and accommodation-only tourists makes it impossibleto separate accommodation and food expenditure meaningfully Othersubdivisions (for example discretionary expenditure on activities non-grocery

29Determinants in tourist expenditure composition

shopping and moving around the destination) are technically feasible but canonly increase the percentages of zeros The CODA methodology involves somedegree of amalgamation of components Second even though there are someadvantages to using a database from an official statistics institution as in thiscase (large sample size and scope) the main disadvantage is that the set ofavailable variables cannot be controlled by the researcher A further issue to beaddressed is the time dimension As the EGATUR survey is conducted annuallyfurther research can be done to include a repeat cross-section analysis in orderto capture trends in the effects of predictors

References

Aitchison J (1986) The Statistical Analysis of Compositional Data Monographs on Statistics and AppliedProbability Chapman and Hall London (Reprinted in 2003 with additional material by TheBlackburn Press)

Alegre J Mateo S and Pou LL (2009) lsquoParticipation in tourism consumption and the intensityof participation an analysis of their socio-demographic and economic determinantsrsquo TourismEconomics Vol 15 No 3 pp 531ndash546

Bewley R (1982) lsquoThe generalized addilog demand system applied to Australian time series andcross section datarsquo Australian Economic Papers Vol 21 No 38 pp 177ndash192

Bewley R and Fiebig DG (1988) lsquoEstimation of price elasticities for an international telephonedemand modelrsquo The Journal of Industrial Economics Vol 36 No 4 pp 393ndash409

Cai LA (1998) lsquoAnalysing household food expenditure patterns on trips and vacations a Tobitmodersquo Journal of Hospitality and Tourism Research Vol 22 No 4 pp 338ndash358

Cai LA (1999) lsquoRelationship of household characteristics and lodging expenditure on leisure tripsrsquoJournal of Hospitality and Leisure Marketing Vol 6 No 2 pp 5ndash18

Cai LA Hong G-S and Morrison AM (1995) lsquoHousehold expenditure patterns for tourismproducts and servicesrsquo Journal of Travel and Tourism Marketing Vol 4 No 4 pp 15ndash39

Cannon TF and Ford J (2002) lsquoRelationship of demographic and trip characteristics to visitorspending an analysis of sports travel visitors across timersquo Tourism Economics Vol 8 No 3 pp 263ndash271

Castillo-Manzano JL and Marchena-Goacutemez M (2010) lsquoAnalysis of determinants of airline choiceprofiling the LCC passengerrsquo Applied Economics Letters Vol 18 No 1 pp 49ndash53

Chiou Y-C and Chen Y-H (2010) lsquoFactors influencing the intentions of passengers regarding full ser-vice and low cost carriers a notersquo Journal of Air Transport Management Vol 16 No 4 pp 226ndash228

Coenen M and van Eekeren L (2003) lsquoA study of the demand for domestic tourism by Swedishhouseholds using a two-staged budgeting modelrsquo Scandinavian Journal of Hospitality and TourismVol 3 No 2 pp 114ndash133

Craggs R and Schofield P (2009) lsquoExpenditure-based segmentation and visitor profiling at theQuays in Salford UKrsquo Tourism Economics Vol 15 No 1 pp 243ndash260

Deaton A and Muellbauer J (1980) lsquoAn almost ideal demand systemrsquo American Economic ReviewVol 70 No 3 pp 312ndash326

Dellaert B Borgers A and Timmermans H (1997) lsquoConsumer activity pattern choice devel-opment and test of stage-dependent conjoint choice experimentsrsquo Journal of Retailing and ConsumerServices Vol 4 No 1 pp 25ndash37

Divisekera S (2007) Modelling and Estimation of Tourism Demand Elasticities A Study of TourismExpenditure Allocation in Australia Cooperative Research Centre for Sustainable Tourism (CRCST)Gold Coast

Divisekera S (2009) lsquoEx-post demand for Australian tourism goods and servicesrsquo Tourism EconomicsVol 15 No 1 pp 153ndash180

Divisekera S (2010) lsquoEconomics of leisure and non-leisure tourist demand a study of domesticdemand for Australian tourismrsquo Tourism Economics Vol 16 No 1 pp 117ndash136

Dolnicar S Crouch GI Devinney T Huybers T Louviere JJ and Oppewal H (2008)lsquoTourism and discretionary income allocation Heterogeneity among householdsrsquo TourismManagement Vol 29 No 1 pp 44ndash52

Downward P and Lumsdon L (2000) lsquoThe demand for day-visits an analysis of visitor spendingrsquoTourism Economics Vol 6 No 3 pp 251ndash261

TOURISM ECONOMICS10

travelling with friends have a larger share of discretionary expensesand highly educated travellers have a larger share of discretionaryexpenses for both airline types

Keywords tourism demand tourist expenditure low-cost airlinecompositional data analysis (CODA)

Independent tourists represent a growing tourist segment Independent touristsdo not travel on a package deal and often organize the entire trip themselvesThey also have easier access to information and usually prefer to undertake moreactivities In 2004 48 of all tourists to Spain travelled without a packagedeal this rose to 66 in 2010 The increase in the number of independenttourists is accompanied by an increasing expansion and consolidation of no-frillsairlines also referred to as low-cost airlines (LCAs) More LCA users than legacyairline passengers travel independently of package tours In 2010 74 of alltourists who arrived in Spain with an LCA did not book a package deal (IET2004 2010)

The cheaper fares offered by LCAs result in a significant reduction intransportation costs which is expected to modify the composition of the triprsquosbudget (Martiacutenez-Garcia and Raya 2008) This seems to have been the casein Spain a major world tourist destination in 2010 557 of all air travellersarrived by LCA and spent 32 of their trip budget on transportation and 54on accommodation and food In comparison legacy passengers spent relativelymore on transportation (40) and somewhat less on accommodation and food(46) (IET 2010) The reduction in transportation costs can also affect thedistribution of non-transportation expenditures in other words at-destinationexpenditures on accommodation activities and other budget components

LCAs have received growing attention in the literature which either focuseson the demand differences between LCA and legacy airlines (OrsquoConnell andWilliams 2005 Chiou and Chen 2010 Forgas et al 2010 Ferrer-Rosell et al2014) or on LCA demand and its heterogeneity (Castillo-Manzano and Marchena-Goacutemez 2010 Martiacutenez-Garcia and Royo-Vela 2010 Kim and Lee 2011Raya-Vilchez and Martiacutenez-Garcia 2011 Martiacutenez-Garcia et al 2012)However this growing literature on LCA demand and comparisons with legacyairline demand has not to our knowledge focused on tourist expenditure andits composition

Expenditure analyses have been quite frequent in the tourism literature sincethey are of major concern for destination management offices (DMO)marketers tourist agencies and in general to all those engaged in tourismTourist expenditure rather than the number of tourists received is becomingmuch more important for destinations and the economic impact of tourism Theanalysis of expenditure composition provides valuable information fordestination management (over that given by the analysis of absoluteexpenditure) in terms of the type of tourist classified by how they distributetheir travel budget The travel budget can be broken down as with householdbudget studies between non-discretionary and discretionary components Non-discretionary components are accommodation transportation and food wherea minimum amount has to be spent whereas discretionary expenditure includes

11Determinants in tourist expenditure composition

extra activities shopping and so on Depending on personal economic trip andsocio-demographic characteristics tourists may be more or less willing toembark on visits activities excursions shopping and so on and thus theproportion of discretionary tourist budget will change accordingly This distinc-tion between expenditure components is relevant for destinations as they arenaturally more interested in local spending than in expenses paid directly totour operators as is the case for package trips and transportation expensesDMOs are also interested in how the different tourist profiles within LCA andlegacy users allocate their budget For instance if DMOs seek to promoteactivities they should focus their marketing efforts on those tourist typesspending more of their budget on that component and which may differbetween airline types

The objective of this article is to study the determinants of the compositionof tourist expenditure in other words the share of tourism expenditureallocated to the different categories of a trip budget by taking into accounttourist heterogeneity and distinguishing between legacy and LCA marketsegments Two main research questions are addressed

bull How does travelling with an LCA versus a legacy airline affect thedistribution of trip expenses between transportation and other costs and thedistribution of the non-transportation (or also called at-destination) expensesbetween discretionary and non-discretionary

bull Are passenger characteristics affecting budget composition If so do theyaffect it in the same way for both types of airline In other words does theairline type have a moderating effect

With this purpose in mind we build a statistical model explaining budgetcomposition from passenger characteristics along with the moderating effect ofairline type The following budget parts are distinguished

bull proportion or share of total trip expenditure devoted to transportation wheresavings from LCAs arise (amount paid for transportation from the airportof origin to the point of accommodation and the return trip)

bull share of total trip expenditure devoted to accommodation and food (basicthat is non-discretionary)

bull share of total trip expenditure devoted to doing activities moving aroundat the destination and shopping (discretionary tourist expenditure)

Statistical analysis of budget compositions is a methodologically challengingtask Shares in budgets as any other composition are expressed as proportionsor percentages of a total whose sum can only be 1 or 100 Compositional datalie in a restricted space and only convey information regarding the relative sizeof components to one another Aitchisonrsquos (1986) seminal work started a fruitfultradition in compositional data analysis (CODA) and of which the most widelyused technique is the transformation of compositions This is achieved by meansof logarithms of ratios Working with log-ratios has not only methodologicalimplications but also substantive ones Without log-ratios components areestimated and interpreted separately from one another as if they could varyindependently (ceteris paribus) which is impossible the relative importance of

TOURISM ECONOMICS12

one component (budget share) can increase only if the relative importance ofat least one other decreases Most methods used to model budget share suchas the almost ideal demand system (Deaton and Muellbauer 1980) ignore atleast partly the constraints and distributional nature of compositional data Tothe best of our knowledge there is no scholarly study of the composition oftourist budget using an appropriate methodology for compositional data analysis

This article is structured as follows First we present a review of the mainapproaches in modelling tourist expenditure which is then followed by adescription of the CODA methodology Then we introduce the method anddata which are followed by the results and finally by the overall conclusionsand discussion

Major approaches in modelling tourism expenditure

The study of expenditure in terms of composition (relative share of each partof the budget) is not the same as the study of expenditure in absolute termsand the variables affecting absolute expenditure may differ from those affectingrelative expenditure Tourism budgets have been approached from bothperspectives in the literature More precisely tourism expenditure has beenanalysed globally in absolute terms per budget parts in relative terms (share)per budget parts and as a part in itself in family budgets

The vast majority of microeconomic tourism demand studies (24 out of 27in the review of Wang and Davidson (2010)) concern the prediction of onesingle aggregated expenditure variable Methods range from mean comparisontests (Craggs and Schofield 2009) OLS and WLS regression (Downward andLumsdon 2000 2003 Cannon and Ford 2002) to advanced econometrictechniques For instance Hung et al (2012) use quantile regression to buildequations predicting not only typical expenditures but also the highest andlowest Alegre et al (2009) Eugenio-Martin (2003) Hong et al (1999) andNicolau and Maacutes (2005) propose double-hurdle Heckit and related models toseparate the decision whether to spend on tourism from the decision of howmuch to spend With respect to the explanatory variables used the mostcommon are income age gender marital status education place of residencelength of stay travel group size and composition accommodation main trippurpose and activities (Marcussen 2011) In general explanatory variables canbe grouped into economic variables (prices and income) socio-demographicvariables and trip or travel-related variables (Wang et al 2006 Sainaghi 2012)Nicolau (2009) includes individual price sensitivity estimated in a previousmodel

Another stream of research is that which analyses tourist expenditure pertourism product (for example lodging food transportation and sightseeingentertainment) A common argument for studying tourist expenditure patternsper tourism product is that it provides vital information to travel organizersand destination marketers when designing the appropriate marketing strategiesFor this purpose researchers have used several methods such as Tobit MANOVAor seemingly unrelated linear regressions (Pyo et al 1991 Cai et al 1995Oppermann 1996 Cai 1998 1999 Lee 2001 Lehto et al 2001 Jang et al2004 Wang et al 2006) Socio-demographic variables are the most common

13Determinants in tourist expenditure composition

significant variables Age affects expenditure on meals as does travel groupwhich also affects transportation expenditure Marital status usually affects foodand accommodation expenditure

Since part expenditure in absolute terms is related to total expenditure acommon finding in these studies is that some of the explanatory variables affectall budget elements roughly equally For example Cai et al (1995) found thatthe higher the level of education the higher the expenditure in all budget partsand Wang et al (2006) conclude similarly about household income Such resultswill never be obtained when analysing budget share

The empirical analysis of budget share both for tourism expenditure andgeneral family budgets commonly involves estimating an almost ideal demandsystem of equations (Deaton and Muellbauer 1980) The almost ideal demandsystem is designed to analyse the interdependences of budget allocations thusovercoming the limitations of single-equation modelling and has receivedmuch attention in the last decade (Song et al 2012) It directly fits shares(composition) as dependent variables in a set of simultaneous regressions Theapproach has focused mainly on estimating price and income elasticities Studiesthat use macroeconomic data include among others those by Divisekera (20072009 2010) Fujii et al (1985) OrsquoHagan and Harrison (1984) Syriopoulos andSinclair (1993) and Wu et al (2011)

The almost ideal demand system has been applied both to data from a givenorigin to multiple destinations (that is ex ante ndash see Li et al 2004 Divisekera2009) or from multiple origins to a given destination which is the case in ourarticle (ex post ndash see Divisekera 2009) In ex post studies it is commonly assumedthat various commodities can be aggregated to broad bundles of productsprovided that prices in a bundle move in parallel and that the utility functionwith respect to tourism and other goods is weakly separable This makes itpossible to conceptualize tourism consumption as a multi-stage process In thefirst stage tourists allocate a household budget part to tourism consumptionin the second a tourism budget part to each tripdestination in the third adestination budget part to each good and service including transportationaccommodation and so on Ex post studies such as ours model only the laststage There are however studies of how tourism competes against othercategories of discretionary expenditure using individual micro data Forexample Melenberg and Van Soest (1996) use different parametric and semi-parametric Tobit models to explain the vacation budget share from householdcharacteristics and Dolnicar et al (2008) analyse how households allocatediscretionary income between tourism and competing uses

The review of Wang and Davidson (2010) concludes that since the vastmajority of tourism demand studies are conducted at macro level (this holdseven more for budget share analysis) there is room for more micro-econometricstudies in this area as the only manner of accounting for demand heterogeneityThe almost ideal demand system approach when applied to micro data caninclude individual characteristics For example Coenen and van Eekeren (2003)and Fleischer et al (2011) make ex ante studies of individual budget share inSweden and Israel respectively and use previous Heckit-type selectionequations to model the decision whether to travel or not Coenen and vanEekeren (2003) include household size and income as individual characteristicsFleischer et al (2011) add age education real state ownership place of birth

TOURISM ECONOMICS14

and Internet use The basic aim of those articles is to estimate elasticities andthey use the individual characteristics only as controls Fleischer et al (2011)are the only researchers to provide the estimates of equations predicting budgetshare from traveller characteristics as a by-product of their elasticity estimatesBeing born in the country of origin is reported to increase the transportationshare and reduce the share of on-site expenditures Education and householdreal estate assets reduce the share of on-site expenditures

The aim and approach of this article are similar to those of Coenen and vanEekeren (2003) and of Fleischer et al (2011) regarding the use of individualcharacteristics to predict budget share but differ in three important respects

bull Our study is ex post so that Heckit modelling is unfeasiblebull The effect of individual characteristics on budget composition (share) is the

core of the analysis Explanatory variables are individual characteristics ratherthan prices The results will include the effects of traveller heterogeneityrather than demand equations

bull Our analysis takes into account the compositional restrictions of the databy using the CODA methodology

The CODA methodology

Compared to absolute data compositional data such as budget share lie in aconstrained space A D-term composition measured on individual i xi1 xi2hellipxiD

has the following constraints

0 le xid le 1 and DΣ

d=1xid = 1 (1)

Aitchison (1986) and Pawlowsky-Glahn and Buccianti (2011) warn against theserious problems that arise when using standard statistical analysis tools oncompositional data Compositional data are non-normal and heteroscedasticOne component can increase only if some other(s) decreases This results innegative spurious correlations among the components and prevents interpretingeffects of linear models in the usual way lsquokeeping everything else constantrsquo

Even if specialized CODA techniques are starting to appear (for exampleRonning 1992 Thioacute-Henestrosa and Martiacuten-Fernaacutendez 2005) the easy way(Aitchison 1986 McLaren et al 1995 Fry et al 1996) involves transformingcompositional data so that they can be subject to standard and well-understoodstatistical techniques This is the approach we take in this article In short thisinvolves using the transformed share by means of logarithms of ratios insteadof the raw share

To ensure compositional coherence of predicted budget share (unit sum andnon-negativity) a set of parameter constraints is imposed to the almost idealdemand system However the presence of an error term with an unboundeddistribution (usually normal) results in a non-zero probability that actual sharelies outside the [01] interval (McLaren et al 1995 Fry et al 1996 Fry 2011)In other words the in fact bounded distribution of budget share results in amisspecification of the almost ideal demand system and of any model fittingpercentage share with an unbounded error distribution

15Determinants in tourist expenditure composition

The fact that the error term in any proper system of demand equationsapplied to budget share should take into account the data compositional naturehas been widely acknowledged (Aitchison 1986 Ronning 1992 Fry et al1996) However the review in Fry (2011) reports few studies that have appliedthe CODA methodology to demand equations and to the study of householdbudgets (McLaren et al 1995 Fry et al 1996 2000)

When fitting demand equations to log-ratios of expenditure componentsEngelrsquos aggregation condition which requires that household total expenditureshould equal the sum of components is automatically satisfied The model caneasily be extended to a full consumer demand analysis by the introduction ofprices and other covariates such as consumer characteristics (Aitchison 1986)McLaren et al (1995) relate the CODA analysis with log-ratio transformationto the almost ideal demand system and conclude that CODA makes it possibleto reach the same objectives with normal and homoscedastic error terms

Related developments are the indirect addilog system (Houthakker 1960)and the generalized addilog system (Bewley 1982) When applied tocompositions (for example Bewley and Fiebig 1988) the latter is equivalentto the CODA methodology in which the log-ratios of each component over thegeometric means of all components are the dependent variables in the set ofsimultaneous regressions

To the best of our knowledge there is no study of tourism budget share usingthe CODA methodology or any other methodology accounting for thecompositional constraints in compositional data

Method and data

Statistical approach

The simplest CODA approach involves applying standard statistical techniqueson logarithms of ratios of components Several log-ratio transformations havebeen suggested in the early CODA literature (Egozcue et al 2003) The additivelog-ratio transformation (alr) used by Fry et al (1996 2000) is the most popularand the easiest to compute given that it is simply the log-ratio of eachcomponent to the last

yid = ln(xidxiD) = ln(xid) ndash ln(xiD) with d = 123D ndash 1 (2)

The centred log-ratio transformation (clr) used by the generalized addilogdemand system (Bewley 1982) computes the log-ratios of each component overthe geometric mean of all the components including itself

⎛ xid ⎞yid = ln ⎜ ndashndashndashndashndashndashndashndashndashndashndashndash ⎟ with d = 123D (3)

⎝ D xi1xi2xi3 xiD ⎠

All log-ratio transformed yid variables recover the full unconstrained ndashinfin to infinrange It must be noted that one dimension is lost in the alr while in the clrone dimension is a linear combination of the remaining

TOURISM ECONOMICS16

The alr is commonly used for statistical modelling and prediction ofcompositions (Fry et al 1996) Conversely the clr transformation is commonlyused for statistical techniques which are based on a metric such as a clusteranalysis because of its preservation of distances even though it leads to asingular covariance matrix Thus while the alr would be appropriate for thepurpose of this article the fact that one component must be used as referencefor all others reduces its flexibility and interpretability Alternatives arepresented at the end of this subsection

While having zero expenditures in absolute data indeed has significantmethodological consequences (Lee 2001) these consequences are arguably moreserious in the CODA methodology If the xid variables contain zeros then log-ratios cannot be computed An obvious initial procedure to reduce zeros is toamalgamate small and conceptually similar components with many zeros intolarger ones In tourism budget research it can be useful to group together allexpenditure on activities or all expenditure on food for instance

In certain instances some zero components result from individual character-istics which are called essential zeros in the CODA literature (Aitchison 1986)Another typology of zeros encountered in the CODA literature is the roundingzero that is a component which is present but is too small to be detected bythe measurement instrument This is typical in chemical biological andgeological compositions and the CODA literature offers ample instruments todeal with rounding zeros

A classic essential zero example in economics is in household budget researchwhen measuring expenditure on tobacco and will essentially be zero if allmembers are non-smokers If tobacco expenditure is decidedly in the researchersrsquointerest the target population should be redefined to include only smokers Inmany instances budget research is often not clear whether zero expenditurescome closer to being essential or rounding zeros In some cases they can beunderstood as corner solutions in a utility maximization problem In others theycan be understood as the inherent randomness of human behaviour or as thelimitations of the data Tourists may spend a certain amount on activities oncertain trips but not on others and so surveys of only one trip will unavoidablycontain some zeros of this type Tourists may also forget or fail to report trivialexpenses such as postcard shopping local bus tickets and going to a museum(see Legoheacuterel (1998) for a discussion on the instability of tourism expenditure)Fry et al (2000) claim that in both situations zeros can be proxied by a verysmall value and thus be treated as rounding zeros In tourist budget researchto treat zero expenses in activities as rounding zeros implies assuming firstthat there is basically no tourist type who will never spend anything onactivities and second that tourists who generally spend little on activities arebasically similar to those who spend nothing or fail to report small expensesWe find both assumptions to be reasonable

Fry et al (2000) essentially used the same zero replacement strategy that waslater suggested by Martiacuten-Fernaacutendez et al (2003) namely replacing xid = 0 with

xprimeid = kδid with 0 lt k lt 1 (4)

where δid is the smallest detectable proportion for individual i and component d

17Determinants in tourist expenditure composition

Martiacuten-Fernaacutendez et al (2003) suggest using k = 065 although a sensitivityanalysis of the results on the choice of k is always advisable (we used k = 030and k = 099 with no sizeable change in the estimates) Next non-zero xid valueshave to be reduced in order to preserve the unit sum and the ratios among non-zero components As suggested by Martiacuten-Fernaacutendez et al (2003) with

xprimeid = xid (1 ndash Σxid=0

xprimeid) (5)

Simulations show this method performs particularly well if the proportion ofzeros is below 10 (Martiacuten-Fernaacutendez et al 2011)

In our data set zeros were present only in one budget category (discretionaryexpenditure) The minimum amount spent by the non-zero group was euro1which roughly corresponds to the price of a city bus ticket the entrance to asubsidized local museum or a cheap souvenir Since the total expenditureis known for each individual we compute δid by dividing euro1 with the totalexpenditure of individual I (see the appendix for the SPSS commandsyntax)

In this article we consider alternative log-ratio transformations which aremore flexible than the alr and clr in that the denominator does not have tobe the same in all ratios This increased flexibility makes it easier to computelog-ratios which are more interpretable with respect to the researchersrsquo questions orhypotheses

In general an interpretable log-ratio transformation is easy to computewhenever there is an interpretable sequential binary partition of componentsinto pairs of groups of components according to the researchersrsquo objectives orto the conceptual similarity of the components These partitions start bydividing components into two clusters and then continue by subdividingone of the clusters into two until each component constitutes its owncluster D components always involve Dndash1 partitions These partitions are bestunderstood as a partition tree or dendrogram (Pawlowsky-Glahn and Egozcue2011)

A meaningful log-ratio transformation takes ratios of the geometric meansof the two component clusters at each partition Numerators and denominatorsare interchangeable In our article we consider xi1 = transportation expenditurexi2 = accommodation and food (basic expenditure) and xi3 = activities andshopping (discretionary expenditure) An interpretable sequential partition isshown in Figure 1

The sequential partition in Figure 1 can have a dual interpretation It mayimply that researchers consider both types of at-destination expenses to bemutually similar and less similar to transportation Or it may imply that theresearch questions involve the distribution of total expenditure betweentransportation and at-destination expenditure and the distribution ofat-destination expenditure into basic and discretionary components as is thecase in our article

The first log-ratio compares transportation expenditure with the geometricmean of accommodation and food (basic expenditure) and activities andshopping (discretionary expenditure) With this ratio we want to observe howtravelling with an LCA or a legacy company affects the share of transportationcompared to non-transportation (at-destination) expenses

TOURISM ECONOMICS18

Figure 1 Sequential partition of total expenditure

⎛ xi1 ⎞ 1 1yi1 = ln ⎜ ndashndashndashndash ⎟ = ln(xi1) ndash ndash ln(xi2) ndash ndash ln(xi3) (6)

⎝ xi2xi3 ⎠ 2 2

Positive values show a transportation share greater than the geometric meanof the remaining two components Negative values show the opposite

The second log-ratio is a ratio of accommodation and food (basic expenditure)over activities and shopping (discretionary expenditure) With this ratio wewant to find out to what tourists allocate the rest of their trip budget andwhether more is allocated to basic or discretionary expenditure once they havepaid for (often in advance) their transportation

⎛ xi2 ⎞yi2 = ln ⎜ ndashndashndash ⎟ = ln(xi2) ndash ln(xi3) (7)

⎝ xi3 ⎠

Positive values show a basic (accommodation and food) share which is largerthan the discretionary (activities and shopping) share Negative values illustratethe opposite

Log-ratio transformations based on sequential binary partitions areproportional to the isometric log-ratio transformation (ilr see Egozcue et al2003 Egozcue and Pawlowsky-Glahn 2005)

As compositions are vector variables they cannot be analysed component-wise by means of univariate regression ANOVA models and the likeSeemingly unrelated regression models for continuous explanatory variables ormultivariate ANOVA (MANOVA) models for categorical explanatory variablesare appropriate MANOVArsquos multivariate tests and statistics (for example

x2 x3 Transportation

Global partition of total expenditure

x1

Accommodation

amp food

(basic)

Activities

amp shopping

(discretionary)

Partition of

expenses other

than transportation

(at-destination

expenses)

19Determinants in tourist expenditure composition

Pillairsquos trace Hotellingrsquos trace or Wilkrsquos lambda) are invariant to howcomponents are arranged in Figure 1 to compute log-ratios (Mateu-Figueras etal 2013) Univariate tests referring to each particular log-ratio are not invari-ant hence the importance of the interpretability of each log-ratio The softwareSPSS 19 is used to estimate the MANOVA model (GLM procedure)

We have to distinguish between three types of variables moderatingexogenous and endogenous to be included in the analysis

Moderating variables are those which modify the effect of exogenousvariables In our article we treat type of airline as a moderator in other wordswe include its interaction terms with all other variables in the MANOVA model

Exogenous variables are assumed to affect expenditure but not the converseSocio-demographic characteristics (travellerrsquos age gender education incomeetc) are obviously determined prior to any travel decision and hence exogenouswith respect to expenditure

Many of the choices tourists make are interdependent (Dellaert et al 1997)or at least planned simultaneously (Fesenmaier and Jeng 2000) So expenditureis arguably decided on at the same time as many trip attributes At least itis unclear whether expenditure is consistently decided on after trip attributesby all travellers We consider them to be endogenous variables and thereforedo not include them in the MANOVA model as explanatory in order to preventendogeneity problems Instead we display them graphically in the log-ratiospace The means of both log-ratios within each endogenous category are thecategory coordinates

The large sample size (see next subsection) makes it possible to use lowp-values Moderating effects with p-values higher than 001 according to eitherPillairsquos trace Hotellingrsquos trace or Wilksrsquo lambda were removed from the modelAll the variables and moderating effects included in the final model aresignificant at 001

Sample and variables

In this article we use secondary official statistics data The data were providedby the Instituto de Estudios Turiacutesticos (IET) an official agency of the Ministryof Industry Energy and Tourism and one which produces the majority oftourism data in Spain The survey is known as the Encuesta de Gasto Turiacutestico(EGATUR) in which tourism expenditure and other tourist information suchas trip information and tourist sociodemographic characteristics are studiedThe EGATUR survey conducted in 27 major Spanish airports in 2010 usedCAPI (computer assisted personal interview) to interview tourists leaving thecountry The sample is non-proportionally stratified by country of residenceairport and month See IET (2012) for further details on the EGATURmethodology

Our universe is a subset of the EGATUR universe which consists ofEuropean leisure visitors arriving by air and spending at least one night inSpain We excluded flights from outside Europe because LCAs mostly operateshort-haul flights We also centred our study on only those trips with one singledestination thus excluding multi-stage trips as the decision process regardingexpenditure composition for these trips is expected to fundamentally differ fromthat of single-stage trips Stays of over 120 days were also excluded

TOURISM ECONOMICS20

For this study we did not consider tourists

bull who have essential zeros in accommodation (tourists who own a house at thedestination or tourists who stay with friends or relatives)

bull who do not decide how much they will spend on certain components(business and study trips)

bull who do not pay for the trip themselves (trips paid for by employers familyfriends prizes offers etc)

bull for whom composition is wholly or partly unobserved (package tourists)

The final sample size was N= 19359From the expenditure variables included in the EGATUR survey database

and used in the model as budget components we first put the amount paidfor transportation (x1) This component has no zeros

Second the amount of money paid for accommodation and food isundistinguishable for full-board half-board or bed and breakfast accommodationTherefore we define a joint accommodation and food component (basic expendi-ture) In this component we include not only the amount paid for consumptionin bars and restaurants but also for buying groceries and everyday products insupermarkets (x2) This component has no zeros

Finally EGATUR provides an aggregated expenditure on activities andshopping (except groceries and everyday products) To this we added theconceptually similar amount paid for moving around the destination (publictransportation andor car rented at the destination) in order to build anactivities and shopping component (x3) This component had 98 zeros whichstresses its discretionary character Zeros were replaced as explained in thestatistical approach subsection

We consider as at-destination (or also called non-transportation) expenses thebasic component (food and accommodation) and the discretionary component(activities and shopping)

Share and log-ratios are described in Table 1 Explanatory variables are thelevel of education income country of residence gender travel groupprofessional status (for pensioners we consider their last professional position)and age The age variable is built as a combination of the original age variableand the variable referring to the individual economic situation This is donein order to have a specific pensioner category taking into account the varyingretirement ages across countries and professions and thus capturing thosetourists who have more free time to undertake a trip regardless of physical ageTable 2 shows their categories and frequencies

As endogenous variables which are not included in the MANOVA modelbut are included in the log-ratio space we include activities undertaken at thedestination accommodation length of stay total expenditure quartilesquartiles of expenditure made at destination per day (basic plus discretionaryexpenses) and a combination of motivation and destination This last variablehas been constructed as a combination of the motivation and the destinationvariable We distinguish between first those tourists travelling for culturaltourism to singular cities second those who come just for leisure either inthe countryside or more commonly at the seaside and third other typologies(Table 3)

21Determinants in tourist expenditure composition

Table 1 Percentage share and log-ratio descriptive statistics

Min Max Mean SD Skewness Kurtosis

Transportation component (x1) 003 9611 2846 1387 0837 0789Basic component (x2) 021 9963 5472 1613 ndash0114 ndash0450Discretionary component (x3) 001 8668 1681 1292 0988 1174Transportationat-destinationlog-ratio (y1) ndash757 548 01506 10702 0714 1404

Basicdiscretionary log-ratio (y2) ndash530 872 17657 18173 1573 2151

Results

Results for the exogenous variables

Table 4 presents the MANOVA results The univariate R2 corrected for thedegrees of freedom is 0045 (first log-ratio) and 0105 (second log-ratio) Themultivariate uncorrected R2 is 1 ndash Λ = 0178 and corrected for the degrees offreedom is 0176 We checked that the results did not change when removingthe four outliers with the highest Cookrsquos distance

Table 4 presents the estimates for the two log-ratios the log-ratio oftransportation over the other two categories (y1) and the log-ratio of basicexpenditure (accommodation and food) over discretionary expenditure (activitiesand shopping) (y2) respectively Because of the moderating effects within eachlog-ratio there are two columns representing the effects when flying with anLCA or with a legacy airline (main and moderating effects have already addedtogether for easier reading) If the LCA and legacy columns are different thereis a significant moderating effect (p-value lt 001)

As we are interpreting log-ratios a positive estimate of a given predictorcategory means that tourists in that category spend more on the numeratorcompared to the denominator than tourists in the reference predictor categoryA negative estimate shows the opposite

The intercept term shows the main effect of company type that is thepredicted log-ratio for each type of airline within the reference category for allvariables (university education medium income resident in the UK or Ireland25ndash44 yearsrsquo old male travelling with partner and mid-level employee) Withinthe reference categories legacy users spend a higher proportion on transportationcompared to other expenses and LCA users spend more on basic expensescompared to discretionary

The level of education seems to have more to do with activities undertakenthan with transportation Results show that a lower level of education resultsin higher expenditure on basic expenses compared to discretionary and is almostequal for users of both types of airline For mainly or only LCA users a lowerlevel of education results in a higher share in transport compared to the othertwo categories If we put it in another way there seems to be a distinct highlyeducated LCA user segment which utilizes the savings in transport toincrease expenditure in non-transportation expenses and even more so indiscretionary

TOURISM ECONOMICS22

Table 2 Frequency distributions of exogenous and moderating variables in the MANOVAmodel

Variables Count Percentage

Level of educationUp to high school 6019 311Universitya 13340 689

Income categoryMediumlow 797 41Mediuma 13186 681Mediumhigh 4376 226High 1000 52

Country of residencePortugal 624 32Other European countries 850 44Belgium 1006 52Netherlands 1069 55Austria Switzerland and Liechtenstein 1122 58France 1326 68Scandinavian countries 1789 92Germany 2363 122Italy 3113 181United Kingdom and Irelanda 6097 315

GenderFemale 8530 441Malea 10829 559

Travel groupAlone 2484 128In family 2980 154With friends 4440 229With partnera 9446 488

Professional statusHomemaker 517 27Unemployed 602 31Student 1432 74Low-level employee 915 47Mid-level employeea 10964 586High-level employee 2138 110Self-employed 2791 144

AgeOver 45 yearsrsquo old and pensioner 1484 77Over 45 yearsrsquo old and not pensioner 5009 25925ndash44 yearsrsquo olda 10569 54615ndash24 yearsrsquo old 2297 119

Type of airline (moderating variable)Legacy 5884 304Low cost 13475 696

Note aReference categories in the MANOVA model chosen either because they are the largestcategories or because they are the most standard conceptually considered

23Determinants in tourist expenditure composition

Table 3 Frequency distributions of endogenous variables (not included in the MANOVAmodel)

Variables Count Percentage

ActivitiesGolfYes 227 12No 19132 988HikingYes 364 19No 18995 981Sporting eventsYes 603 31No 18756 969Nautical sportsYes 1645 85No 17714 915Other sportsYes 904 47No 18455 953Cultural visitsYes 11180 578No 8179 422Cultural eventsYes 2452 127No 16907 873Other cultural activitiesYes 4175 216No 15184 784SpaYes 1199 62No 18160 938Theme parksYes 1745 90No 17614 910GastronomyYes 1565 81No 17794 919Type of accommodationHotel 4ndash5 5056 261Hotel 3 4922 254Hotel lt 3 6673 345Rented appartment 2116 109Other accommodation 592 31Length of stay5 or less nights 9911 5126ndash8 nights 5952 3079ndash12 nights 1592 8213ndash15 nights 1261 6516 and more nights 643 33

TOURISM ECONOMICS24

Table 3 continued

Variables Count Percentage

Total expenditureQuartile 1 5074 262Quartile 2 5104 264Quartile 3 4986 258Quartile 4 4195 217Expenditure at destination per dayQuartile 1 4855 251Quartile 2 4832 250Quartile 3 4868 251Quartile 4 4804 248Motivation-destinationLeisure on coastin countryside 8083 418Urban tourism 5324 275Others 5952 307

Level of income in contrast affects both log-ratios The higher the incomethe lower the transport share The higher the income the lower the basicdiscretionary expense ratio that is a higher portion of the non-transportexpense is devoted to the discretionary budget

As far as the country of residence is concerned the results show that it affectsmainly the basicdiscretionary log-ratio and exhibits strong airline typemoderating effects LCA users tend to differ depending on their country ofresidence when distributing non-transport expenses into basic versusdiscretionary In general when compared to the UK and Ireland residentcategory (reference category) all other European residents have a largerdiscretionary share More precisely the French Italians and Portuguese have thehighest discretionary share within LCA users and other European countries(mainly central and eastern European) have the highest discretionary sharewithin legacy airline users

Gender has a small effect which is constant for both types of airline Femaletravellers tend to slightly increase the transport share and the share of basicexpenses within the non-transport expenses

The travel group seems to affect both log-ratios and has a moderating effectwith airline type Those tourists who travel in a family group have a highershare in transport expenses when flying with legacy airlines In the second log-ratio travelling as a family with legacy airlines increases the basic expensesshare compared to the share in discretionary Travelling with friends increasesthe share in non-transport expenses for both airline types Besides those whotravel with friends spend more in the discretionary share compared to basicexpenses and those who fly with LCA even more so Finally travelling aloneincreases the basic expenses share compared to the share in discretionaryexpenses for both airline types

When professional status is considered the results show that it mostly affectsthe basicdiscretionary log-ratio and there are some relevant airline type

25Determinants in tourist expenditure composition

Table 4 MANOVA results

y1 log-ratio y2 log-ratiotransportation basicat-destination discretionary

LCA Legacy LCA Legacy

Intercept 0225 0274 2195 1881

Up to high school 0192 0078 0406 0437University 0 0 0 0

Mediumlow income 0073 0289 0037 0518Medium income 0 0 0 0Mediumhigh income ndash0227 ndash0058 ndash0269 ndash0006High income ndash0495 ndash0190 ndash0463 ndash0141

Residence in Portugal 0041 0069 ndash1136 ndash0580Residence in other European Countries 0326 ndash0012 ndash0772 ndash0820Residence in Belgium ndash0127 ndash0159 ndash0667 ndash0461Residence in Netherlands ndash0120 ndash0197 ndash0678 ndash0530Residence in Austria + Switzerland +Liechtenstein ndash0017 ndash0087 ndash0687 ndash0436

Residence in France ndash0112 ndash0111 ndash0870 ndash0553Residence in Scandinavian countries ndash0045 ndash0038 ndash0660 ndash0477Residence in Germany ndash0283 ndash0176 ndash0452 ndash0392Residence in Italy ndash0209 ndash0266 ndash0855 ndash0503Residence in UK or Ireland 0 0 0 0

Female 0094 0094 0093 0093Male 0 0 0 0

Travelling alone ndash0053 ndash0002 0081 0126Travelling in family 0085 0193 0014 0213Travelling with friends ndash0237 ndash0191 ndash0504 ndash0282Travelling with partner 0 0 0 0

Homemaker 0080 0160 0253 0235Unemployed 0030 0159 ndash0036 ndash0039Student ndash0005 ndash0104 ndash0331 ndash0624Low-level employee ndash0034 0194 ndash0194 0063Mid-level employee 0 0 0 0High-level employee ndash0008 ndash0064 ndash0001 ndash0139Self-employed 0029 ndash0086 0193 ndash0072

Over 45 yearsrsquo old and pensioner ndash0060 ndash0060 0633 0633Over 45 yearsrsquo old and not pensioner ndash0010 ndash0010 0142 014225ndash44 yearsrsquo old 0 0 0 015ndash24 yearsrsquo old 0002 0002 ndash0019 ndash0019

moderating effects Legacy users with a low professional status (homemakerunemployed or low-level employee) spend more on transport compared to thenon-transport expenses Conversely students using legacy airlines spend moreon non-transport expenses and these non-transport expenses are to a greater

TOURISM ECONOMICS26

Figure 2 Endogenous variables on the log-ratio space Mean log-ratio valueswithin each category of the endogenous variables

extent devoted to discretionary expenses compared to basic expensesHomemakers increase basic expenditure compared to discretionary with respectto the reference category (mid-level employee) for both airline types LCA userswho are self-employed also increase basic expenditure compared to discretionaryLow-level employees flying with an LCA increase the discretionary sharecompared to the basic and high-level employees flying with legacy airlines alsospend relatively more on the discretionary part

The results show that age affects the second log-ratio more than the firstand there is no moderating effect with the airline type Being a pensioner resultsin a higher share in basic expenses compared to discretionary expenses for bothairline types The remaining categories have hardly any differences from thereference 25ndash44 yearsrsquo old

Results for the endogenous variables

As introduced in the methodology section Figure 2 shows how the endogenousvariables behave in the log-ratio space The origin of the graph is representedby the mean log-ratios in Table 1 01506 for the transportother ratio and17657 for the basicdiscretionary ratio and according to this the graph showscategories with higher or lower ratios than the average

075

060

045

030

015

000

ndash015

ndash030

ndash045

ndash060

Ratio tra

nsport

ationa

t-destination e

xpenses

Ratio basicdiscretionary expenses

050 075 100 125 150 175 200 225 250

27Determinants in tourist expenditure composition

Regarding accommodation it seems that rented apartments and other typesof accommodation behave similarly and have the lowest budget share intransport In the case of hotels as the star rating increases expenditure on basiccompared to discretionary increases and expenditure on transport compared toat-destination decreases

As far as the length of stay is concerned longer stays increase the share ofnon-transport expenditure in basic expenses and reduce the share in transport

As regards destination and motivation tourists going to a seaside or country-side destination for leisure purposes behave quite differently from those goingto an urban destination for urban or cultural tourism The former group spendsmore on transport and less on the discretionary portion However those comingfor urban tourism show an increase in the share of non-transport expenses aswell as in the discretionary share compared to basic expenses

The total expenditure variable gives relevant information It is ordered alongthe first log-ratio axis from top to bottom meaning that those tourists havinga lower total expenditure spend relatively more on the transport share and thosewho have a higher total expenditure spend more on the other two parts of thebudget Regarding the expenditure made at destination per day it is ordereddiagonally from the upper-right corner to the bottom-left corner meaning thattourists with a higher at-destination expenditure per day spend less ontransportation and at the same time more on the discretionary part within at-destination expenses

Finally activities undertaken are located in the lower left quadrant of thegraph meaning that undertaking activities (discretionary expenses) decreasesthe share in transport expenses and although they continue to spend more onbasic expenses than on discretionary the discretionary share increases We havehighlighted some of them Those tourists attending sports and cultural eventsare those spending the most on discretionary compared to basic expensesHiking or some nautical sports could be free of charge thus tourists doing theseactivities have a basic to discretionary ratio close to the mean As for thetransportation share all activities tend to reduce this share about equally withthe exception of golf which leads to a much more substantial reduction

Conclusion and discussion

The main purpose of this article was to study the composition of touristexpenditure and its determinants that is the drivers of the share of touristexpenditure allocated to the different categories of a travel budget and withspecial emphasis on the distinction between legacy versus LCA travellers

The main differences between legacy and LCA users are that income somecountries of residence the travel group and some occupations have differenteffects depending on the type of airline Tourists with a medium or low incomewho travel with legacy airlines tend to spend a greater share of their travelbudget on transportation and basic expenses but those travelling with LCAstend to spend more at the destination and more specifically ondiscretionary expenditure LCA tourists residing in the lsquoother Europeancountriesrsquo spend more on transportation Families flying with legacy airlinesspend more on the transportation share compared to the at-destination share

TOURISM ECONOMICS28

and also spend more on basic expenses compared to discretionary ones Finallyin terms of occupational status legacy users with low-level employment spendmore on transportation whereas the opposite occurs with LCA travellers whospend more on the discretionary component Students and high-level employeesflying with legacy airlines spend more on the discretionary part of the tripbudget

As far as implications for management are concerned assuming that theobjectives of DMOs are to increase the allocation to the at-destinationcomponent of the trip budget (that is lowering the first log-ratio) some of therecommended actions would be to increase marketing efforts directed atmedium- and high-income earners especially at those travelling on LCAs toincrease marketing strategies in some of the European outbound markets suchas Germany (especially those using LCA) and Italy and the Netherlands(especially those flying with legacy airlines) DMOs should also focus on thosetourists travelling with friends (airline type is inconsequential) and studentsusing legacy companies As for the endogenous variables DMOs should directtheir efforts towards those tourists undertaking activities towards those stayinga little bit longer than a week and towards those with high total expenditureFurthermore if DMOs are also interested in capturing the groups who spendrelatively more on discretionary expenses (lowering the second log-ratio) theyshould focus their marketing efforts on those tourists who are highly educatedas well as those who have medium to high incomes and who use LCAsAdditionally they should focus on markets other than the UK and IrelandYoung tourists and tourists who travel with friends (where airline type isinconsequential) low-level employees and students using LCAs as well asstudents and high-level employees flying on legacy airlines should also betargeted In relation to the endogenous variables DMOs must take into accountthose tourists undertaking payable activities those on shorter stays or comingfor urban and cultural tourism or other purposes

The appeal of the CODA methodology for studying tourism budgets lies inthe fact that once the variables have been transformed the researcher can usestandard and well understood statistical models with unbounded error termdistributions while ensuring that the results will be compositionally coherentand that the standard statistical assumptions will hold Conversely results arenot compositionally coherent when we use the MANOVA model on raw sharewithout transforming it into log-ratios We have seen with our data that 84of individuals included in the sample have at least one limit of the 90 shareprediction intervals outside the [01] range

Contrary to the classic and highly restrictive alr and clr transformations theCODA methodology offers the potential to construct tailor-made log-ratios whichare intuitive to interpret and suit the research questions at hand A classificationtree of components is a clear and useful tool in this respect We encourageresearchers to use this approach in further research on tourism expenditure whenbudget share division is fundamental to the researchersrsquo questions

In this study we have encountered two limitations First contrary to studiesthat use macro data in micro data studies the coexistence of full-board halfboard bed and breakfast and accommodation-only tourists makes it impossibleto separate accommodation and food expenditure meaningfully Othersubdivisions (for example discretionary expenditure on activities non-grocery

29Determinants in tourist expenditure composition

shopping and moving around the destination) are technically feasible but canonly increase the percentages of zeros The CODA methodology involves somedegree of amalgamation of components Second even though there are someadvantages to using a database from an official statistics institution as in thiscase (large sample size and scope) the main disadvantage is that the set ofavailable variables cannot be controlled by the researcher A further issue to beaddressed is the time dimension As the EGATUR survey is conducted annuallyfurther research can be done to include a repeat cross-section analysis in orderto capture trends in the effects of predictors

References

Aitchison J (1986) The Statistical Analysis of Compositional Data Monographs on Statistics and AppliedProbability Chapman and Hall London (Reprinted in 2003 with additional material by TheBlackburn Press)

Alegre J Mateo S and Pou LL (2009) lsquoParticipation in tourism consumption and the intensityof participation an analysis of their socio-demographic and economic determinantsrsquo TourismEconomics Vol 15 No 3 pp 531ndash546

Bewley R (1982) lsquoThe generalized addilog demand system applied to Australian time series andcross section datarsquo Australian Economic Papers Vol 21 No 38 pp 177ndash192

Bewley R and Fiebig DG (1988) lsquoEstimation of price elasticities for an international telephonedemand modelrsquo The Journal of Industrial Economics Vol 36 No 4 pp 393ndash409

Cai LA (1998) lsquoAnalysing household food expenditure patterns on trips and vacations a Tobitmodersquo Journal of Hospitality and Tourism Research Vol 22 No 4 pp 338ndash358

Cai LA (1999) lsquoRelationship of household characteristics and lodging expenditure on leisure tripsrsquoJournal of Hospitality and Leisure Marketing Vol 6 No 2 pp 5ndash18

Cai LA Hong G-S and Morrison AM (1995) lsquoHousehold expenditure patterns for tourismproducts and servicesrsquo Journal of Travel and Tourism Marketing Vol 4 No 4 pp 15ndash39

Cannon TF and Ford J (2002) lsquoRelationship of demographic and trip characteristics to visitorspending an analysis of sports travel visitors across timersquo Tourism Economics Vol 8 No 3 pp 263ndash271

Castillo-Manzano JL and Marchena-Goacutemez M (2010) lsquoAnalysis of determinants of airline choiceprofiling the LCC passengerrsquo Applied Economics Letters Vol 18 No 1 pp 49ndash53

Chiou Y-C and Chen Y-H (2010) lsquoFactors influencing the intentions of passengers regarding full ser-vice and low cost carriers a notersquo Journal of Air Transport Management Vol 16 No 4 pp 226ndash228

Coenen M and van Eekeren L (2003) lsquoA study of the demand for domestic tourism by Swedishhouseholds using a two-staged budgeting modelrsquo Scandinavian Journal of Hospitality and TourismVol 3 No 2 pp 114ndash133

Craggs R and Schofield P (2009) lsquoExpenditure-based segmentation and visitor profiling at theQuays in Salford UKrsquo Tourism Economics Vol 15 No 1 pp 243ndash260

Deaton A and Muellbauer J (1980) lsquoAn almost ideal demand systemrsquo American Economic ReviewVol 70 No 3 pp 312ndash326

Dellaert B Borgers A and Timmermans H (1997) lsquoConsumer activity pattern choice devel-opment and test of stage-dependent conjoint choice experimentsrsquo Journal of Retailing and ConsumerServices Vol 4 No 1 pp 25ndash37

Divisekera S (2007) Modelling and Estimation of Tourism Demand Elasticities A Study of TourismExpenditure Allocation in Australia Cooperative Research Centre for Sustainable Tourism (CRCST)Gold Coast

Divisekera S (2009) lsquoEx-post demand for Australian tourism goods and servicesrsquo Tourism EconomicsVol 15 No 1 pp 153ndash180

Divisekera S (2010) lsquoEconomics of leisure and non-leisure tourist demand a study of domesticdemand for Australian tourismrsquo Tourism Economics Vol 16 No 1 pp 117ndash136

Dolnicar S Crouch GI Devinney T Huybers T Louviere JJ and Oppewal H (2008)lsquoTourism and discretionary income allocation Heterogeneity among householdsrsquo TourismManagement Vol 29 No 1 pp 44ndash52

Downward P and Lumsdon L (2000) lsquoThe demand for day-visits an analysis of visitor spendingrsquoTourism Economics Vol 6 No 3 pp 251ndash261

11Determinants in tourist expenditure composition

extra activities shopping and so on Depending on personal economic trip andsocio-demographic characteristics tourists may be more or less willing toembark on visits activities excursions shopping and so on and thus theproportion of discretionary tourist budget will change accordingly This distinc-tion between expenditure components is relevant for destinations as they arenaturally more interested in local spending than in expenses paid directly totour operators as is the case for package trips and transportation expensesDMOs are also interested in how the different tourist profiles within LCA andlegacy users allocate their budget For instance if DMOs seek to promoteactivities they should focus their marketing efforts on those tourist typesspending more of their budget on that component and which may differbetween airline types

The objective of this article is to study the determinants of the compositionof tourist expenditure in other words the share of tourism expenditureallocated to the different categories of a trip budget by taking into accounttourist heterogeneity and distinguishing between legacy and LCA marketsegments Two main research questions are addressed

bull How does travelling with an LCA versus a legacy airline affect thedistribution of trip expenses between transportation and other costs and thedistribution of the non-transportation (or also called at-destination) expensesbetween discretionary and non-discretionary

bull Are passenger characteristics affecting budget composition If so do theyaffect it in the same way for both types of airline In other words does theairline type have a moderating effect

With this purpose in mind we build a statistical model explaining budgetcomposition from passenger characteristics along with the moderating effect ofairline type The following budget parts are distinguished

bull proportion or share of total trip expenditure devoted to transportation wheresavings from LCAs arise (amount paid for transportation from the airportof origin to the point of accommodation and the return trip)

bull share of total trip expenditure devoted to accommodation and food (basicthat is non-discretionary)

bull share of total trip expenditure devoted to doing activities moving aroundat the destination and shopping (discretionary tourist expenditure)

Statistical analysis of budget compositions is a methodologically challengingtask Shares in budgets as any other composition are expressed as proportionsor percentages of a total whose sum can only be 1 or 100 Compositional datalie in a restricted space and only convey information regarding the relative sizeof components to one another Aitchisonrsquos (1986) seminal work started a fruitfultradition in compositional data analysis (CODA) and of which the most widelyused technique is the transformation of compositions This is achieved by meansof logarithms of ratios Working with log-ratios has not only methodologicalimplications but also substantive ones Without log-ratios components areestimated and interpreted separately from one another as if they could varyindependently (ceteris paribus) which is impossible the relative importance of

TOURISM ECONOMICS12

one component (budget share) can increase only if the relative importance ofat least one other decreases Most methods used to model budget share suchas the almost ideal demand system (Deaton and Muellbauer 1980) ignore atleast partly the constraints and distributional nature of compositional data Tothe best of our knowledge there is no scholarly study of the composition oftourist budget using an appropriate methodology for compositional data analysis

This article is structured as follows First we present a review of the mainapproaches in modelling tourist expenditure which is then followed by adescription of the CODA methodology Then we introduce the method anddata which are followed by the results and finally by the overall conclusionsand discussion

Major approaches in modelling tourism expenditure

The study of expenditure in terms of composition (relative share of each partof the budget) is not the same as the study of expenditure in absolute termsand the variables affecting absolute expenditure may differ from those affectingrelative expenditure Tourism budgets have been approached from bothperspectives in the literature More precisely tourism expenditure has beenanalysed globally in absolute terms per budget parts in relative terms (share)per budget parts and as a part in itself in family budgets

The vast majority of microeconomic tourism demand studies (24 out of 27in the review of Wang and Davidson (2010)) concern the prediction of onesingle aggregated expenditure variable Methods range from mean comparisontests (Craggs and Schofield 2009) OLS and WLS regression (Downward andLumsdon 2000 2003 Cannon and Ford 2002) to advanced econometrictechniques For instance Hung et al (2012) use quantile regression to buildequations predicting not only typical expenditures but also the highest andlowest Alegre et al (2009) Eugenio-Martin (2003) Hong et al (1999) andNicolau and Maacutes (2005) propose double-hurdle Heckit and related models toseparate the decision whether to spend on tourism from the decision of howmuch to spend With respect to the explanatory variables used the mostcommon are income age gender marital status education place of residencelength of stay travel group size and composition accommodation main trippurpose and activities (Marcussen 2011) In general explanatory variables canbe grouped into economic variables (prices and income) socio-demographicvariables and trip or travel-related variables (Wang et al 2006 Sainaghi 2012)Nicolau (2009) includes individual price sensitivity estimated in a previousmodel

Another stream of research is that which analyses tourist expenditure pertourism product (for example lodging food transportation and sightseeingentertainment) A common argument for studying tourist expenditure patternsper tourism product is that it provides vital information to travel organizersand destination marketers when designing the appropriate marketing strategiesFor this purpose researchers have used several methods such as Tobit MANOVAor seemingly unrelated linear regressions (Pyo et al 1991 Cai et al 1995Oppermann 1996 Cai 1998 1999 Lee 2001 Lehto et al 2001 Jang et al2004 Wang et al 2006) Socio-demographic variables are the most common

13Determinants in tourist expenditure composition

significant variables Age affects expenditure on meals as does travel groupwhich also affects transportation expenditure Marital status usually affects foodand accommodation expenditure

Since part expenditure in absolute terms is related to total expenditure acommon finding in these studies is that some of the explanatory variables affectall budget elements roughly equally For example Cai et al (1995) found thatthe higher the level of education the higher the expenditure in all budget partsand Wang et al (2006) conclude similarly about household income Such resultswill never be obtained when analysing budget share

The empirical analysis of budget share both for tourism expenditure andgeneral family budgets commonly involves estimating an almost ideal demandsystem of equations (Deaton and Muellbauer 1980) The almost ideal demandsystem is designed to analyse the interdependences of budget allocations thusovercoming the limitations of single-equation modelling and has receivedmuch attention in the last decade (Song et al 2012) It directly fits shares(composition) as dependent variables in a set of simultaneous regressions Theapproach has focused mainly on estimating price and income elasticities Studiesthat use macroeconomic data include among others those by Divisekera (20072009 2010) Fujii et al (1985) OrsquoHagan and Harrison (1984) Syriopoulos andSinclair (1993) and Wu et al (2011)

The almost ideal demand system has been applied both to data from a givenorigin to multiple destinations (that is ex ante ndash see Li et al 2004 Divisekera2009) or from multiple origins to a given destination which is the case in ourarticle (ex post ndash see Divisekera 2009) In ex post studies it is commonly assumedthat various commodities can be aggregated to broad bundles of productsprovided that prices in a bundle move in parallel and that the utility functionwith respect to tourism and other goods is weakly separable This makes itpossible to conceptualize tourism consumption as a multi-stage process In thefirst stage tourists allocate a household budget part to tourism consumptionin the second a tourism budget part to each tripdestination in the third adestination budget part to each good and service including transportationaccommodation and so on Ex post studies such as ours model only the laststage There are however studies of how tourism competes against othercategories of discretionary expenditure using individual micro data Forexample Melenberg and Van Soest (1996) use different parametric and semi-parametric Tobit models to explain the vacation budget share from householdcharacteristics and Dolnicar et al (2008) analyse how households allocatediscretionary income between tourism and competing uses

The review of Wang and Davidson (2010) concludes that since the vastmajority of tourism demand studies are conducted at macro level (this holdseven more for budget share analysis) there is room for more micro-econometricstudies in this area as the only manner of accounting for demand heterogeneityThe almost ideal demand system approach when applied to micro data caninclude individual characteristics For example Coenen and van Eekeren (2003)and Fleischer et al (2011) make ex ante studies of individual budget share inSweden and Israel respectively and use previous Heckit-type selectionequations to model the decision whether to travel or not Coenen and vanEekeren (2003) include household size and income as individual characteristicsFleischer et al (2011) add age education real state ownership place of birth

TOURISM ECONOMICS14

and Internet use The basic aim of those articles is to estimate elasticities andthey use the individual characteristics only as controls Fleischer et al (2011)are the only researchers to provide the estimates of equations predicting budgetshare from traveller characteristics as a by-product of their elasticity estimatesBeing born in the country of origin is reported to increase the transportationshare and reduce the share of on-site expenditures Education and householdreal estate assets reduce the share of on-site expenditures

The aim and approach of this article are similar to those of Coenen and vanEekeren (2003) and of Fleischer et al (2011) regarding the use of individualcharacteristics to predict budget share but differ in three important respects

bull Our study is ex post so that Heckit modelling is unfeasiblebull The effect of individual characteristics on budget composition (share) is the

core of the analysis Explanatory variables are individual characteristics ratherthan prices The results will include the effects of traveller heterogeneityrather than demand equations

bull Our analysis takes into account the compositional restrictions of the databy using the CODA methodology

The CODA methodology

Compared to absolute data compositional data such as budget share lie in aconstrained space A D-term composition measured on individual i xi1 xi2hellipxiD

has the following constraints

0 le xid le 1 and DΣ

d=1xid = 1 (1)

Aitchison (1986) and Pawlowsky-Glahn and Buccianti (2011) warn against theserious problems that arise when using standard statistical analysis tools oncompositional data Compositional data are non-normal and heteroscedasticOne component can increase only if some other(s) decreases This results innegative spurious correlations among the components and prevents interpretingeffects of linear models in the usual way lsquokeeping everything else constantrsquo

Even if specialized CODA techniques are starting to appear (for exampleRonning 1992 Thioacute-Henestrosa and Martiacuten-Fernaacutendez 2005) the easy way(Aitchison 1986 McLaren et al 1995 Fry et al 1996) involves transformingcompositional data so that they can be subject to standard and well-understoodstatistical techniques This is the approach we take in this article In short thisinvolves using the transformed share by means of logarithms of ratios insteadof the raw share

To ensure compositional coherence of predicted budget share (unit sum andnon-negativity) a set of parameter constraints is imposed to the almost idealdemand system However the presence of an error term with an unboundeddistribution (usually normal) results in a non-zero probability that actual sharelies outside the [01] interval (McLaren et al 1995 Fry et al 1996 Fry 2011)In other words the in fact bounded distribution of budget share results in amisspecification of the almost ideal demand system and of any model fittingpercentage share with an unbounded error distribution

15Determinants in tourist expenditure composition

The fact that the error term in any proper system of demand equationsapplied to budget share should take into account the data compositional naturehas been widely acknowledged (Aitchison 1986 Ronning 1992 Fry et al1996) However the review in Fry (2011) reports few studies that have appliedthe CODA methodology to demand equations and to the study of householdbudgets (McLaren et al 1995 Fry et al 1996 2000)

When fitting demand equations to log-ratios of expenditure componentsEngelrsquos aggregation condition which requires that household total expenditureshould equal the sum of components is automatically satisfied The model caneasily be extended to a full consumer demand analysis by the introduction ofprices and other covariates such as consumer characteristics (Aitchison 1986)McLaren et al (1995) relate the CODA analysis with log-ratio transformationto the almost ideal demand system and conclude that CODA makes it possibleto reach the same objectives with normal and homoscedastic error terms

Related developments are the indirect addilog system (Houthakker 1960)and the generalized addilog system (Bewley 1982) When applied tocompositions (for example Bewley and Fiebig 1988) the latter is equivalentto the CODA methodology in which the log-ratios of each component over thegeometric means of all components are the dependent variables in the set ofsimultaneous regressions

To the best of our knowledge there is no study of tourism budget share usingthe CODA methodology or any other methodology accounting for thecompositional constraints in compositional data

Method and data

Statistical approach

The simplest CODA approach involves applying standard statistical techniqueson logarithms of ratios of components Several log-ratio transformations havebeen suggested in the early CODA literature (Egozcue et al 2003) The additivelog-ratio transformation (alr) used by Fry et al (1996 2000) is the most popularand the easiest to compute given that it is simply the log-ratio of eachcomponent to the last

yid = ln(xidxiD) = ln(xid) ndash ln(xiD) with d = 123D ndash 1 (2)

The centred log-ratio transformation (clr) used by the generalized addilogdemand system (Bewley 1982) computes the log-ratios of each component overthe geometric mean of all the components including itself

⎛ xid ⎞yid = ln ⎜ ndashndashndashndashndashndashndashndashndashndashndashndash ⎟ with d = 123D (3)

⎝ D xi1xi2xi3 xiD ⎠

All log-ratio transformed yid variables recover the full unconstrained ndashinfin to infinrange It must be noted that one dimension is lost in the alr while in the clrone dimension is a linear combination of the remaining

TOURISM ECONOMICS16

The alr is commonly used for statistical modelling and prediction ofcompositions (Fry et al 1996) Conversely the clr transformation is commonlyused for statistical techniques which are based on a metric such as a clusteranalysis because of its preservation of distances even though it leads to asingular covariance matrix Thus while the alr would be appropriate for thepurpose of this article the fact that one component must be used as referencefor all others reduces its flexibility and interpretability Alternatives arepresented at the end of this subsection

While having zero expenditures in absolute data indeed has significantmethodological consequences (Lee 2001) these consequences are arguably moreserious in the CODA methodology If the xid variables contain zeros then log-ratios cannot be computed An obvious initial procedure to reduce zeros is toamalgamate small and conceptually similar components with many zeros intolarger ones In tourism budget research it can be useful to group together allexpenditure on activities or all expenditure on food for instance

In certain instances some zero components result from individual character-istics which are called essential zeros in the CODA literature (Aitchison 1986)Another typology of zeros encountered in the CODA literature is the roundingzero that is a component which is present but is too small to be detected bythe measurement instrument This is typical in chemical biological andgeological compositions and the CODA literature offers ample instruments todeal with rounding zeros

A classic essential zero example in economics is in household budget researchwhen measuring expenditure on tobacco and will essentially be zero if allmembers are non-smokers If tobacco expenditure is decidedly in the researchersrsquointerest the target population should be redefined to include only smokers Inmany instances budget research is often not clear whether zero expenditurescome closer to being essential or rounding zeros In some cases they can beunderstood as corner solutions in a utility maximization problem In others theycan be understood as the inherent randomness of human behaviour or as thelimitations of the data Tourists may spend a certain amount on activities oncertain trips but not on others and so surveys of only one trip will unavoidablycontain some zeros of this type Tourists may also forget or fail to report trivialexpenses such as postcard shopping local bus tickets and going to a museum(see Legoheacuterel (1998) for a discussion on the instability of tourism expenditure)Fry et al (2000) claim that in both situations zeros can be proxied by a verysmall value and thus be treated as rounding zeros In tourist budget researchto treat zero expenses in activities as rounding zeros implies assuming firstthat there is basically no tourist type who will never spend anything onactivities and second that tourists who generally spend little on activities arebasically similar to those who spend nothing or fail to report small expensesWe find both assumptions to be reasonable

Fry et al (2000) essentially used the same zero replacement strategy that waslater suggested by Martiacuten-Fernaacutendez et al (2003) namely replacing xid = 0 with

xprimeid = kδid with 0 lt k lt 1 (4)

where δid is the smallest detectable proportion for individual i and component d

17Determinants in tourist expenditure composition

Martiacuten-Fernaacutendez et al (2003) suggest using k = 065 although a sensitivityanalysis of the results on the choice of k is always advisable (we used k = 030and k = 099 with no sizeable change in the estimates) Next non-zero xid valueshave to be reduced in order to preserve the unit sum and the ratios among non-zero components As suggested by Martiacuten-Fernaacutendez et al (2003) with

xprimeid = xid (1 ndash Σxid=0

xprimeid) (5)

Simulations show this method performs particularly well if the proportion ofzeros is below 10 (Martiacuten-Fernaacutendez et al 2011)

In our data set zeros were present only in one budget category (discretionaryexpenditure) The minimum amount spent by the non-zero group was euro1which roughly corresponds to the price of a city bus ticket the entrance to asubsidized local museum or a cheap souvenir Since the total expenditureis known for each individual we compute δid by dividing euro1 with the totalexpenditure of individual I (see the appendix for the SPSS commandsyntax)

In this article we consider alternative log-ratio transformations which aremore flexible than the alr and clr in that the denominator does not have tobe the same in all ratios This increased flexibility makes it easier to computelog-ratios which are more interpretable with respect to the researchersrsquo questions orhypotheses

In general an interpretable log-ratio transformation is easy to computewhenever there is an interpretable sequential binary partition of componentsinto pairs of groups of components according to the researchersrsquo objectives orto the conceptual similarity of the components These partitions start bydividing components into two clusters and then continue by subdividingone of the clusters into two until each component constitutes its owncluster D components always involve Dndash1 partitions These partitions are bestunderstood as a partition tree or dendrogram (Pawlowsky-Glahn and Egozcue2011)

A meaningful log-ratio transformation takes ratios of the geometric meansof the two component clusters at each partition Numerators and denominatorsare interchangeable In our article we consider xi1 = transportation expenditurexi2 = accommodation and food (basic expenditure) and xi3 = activities andshopping (discretionary expenditure) An interpretable sequential partition isshown in Figure 1

The sequential partition in Figure 1 can have a dual interpretation It mayimply that researchers consider both types of at-destination expenses to bemutually similar and less similar to transportation Or it may imply that theresearch questions involve the distribution of total expenditure betweentransportation and at-destination expenditure and the distribution ofat-destination expenditure into basic and discretionary components as is thecase in our article

The first log-ratio compares transportation expenditure with the geometricmean of accommodation and food (basic expenditure) and activities andshopping (discretionary expenditure) With this ratio we want to observe howtravelling with an LCA or a legacy company affects the share of transportationcompared to non-transportation (at-destination) expenses

TOURISM ECONOMICS18

Figure 1 Sequential partition of total expenditure

⎛ xi1 ⎞ 1 1yi1 = ln ⎜ ndashndashndashndash ⎟ = ln(xi1) ndash ndash ln(xi2) ndash ndash ln(xi3) (6)

⎝ xi2xi3 ⎠ 2 2

Positive values show a transportation share greater than the geometric meanof the remaining two components Negative values show the opposite

The second log-ratio is a ratio of accommodation and food (basic expenditure)over activities and shopping (discretionary expenditure) With this ratio wewant to find out to what tourists allocate the rest of their trip budget andwhether more is allocated to basic or discretionary expenditure once they havepaid for (often in advance) their transportation

⎛ xi2 ⎞yi2 = ln ⎜ ndashndashndash ⎟ = ln(xi2) ndash ln(xi3) (7)

⎝ xi3 ⎠

Positive values show a basic (accommodation and food) share which is largerthan the discretionary (activities and shopping) share Negative values illustratethe opposite

Log-ratio transformations based on sequential binary partitions areproportional to the isometric log-ratio transformation (ilr see Egozcue et al2003 Egozcue and Pawlowsky-Glahn 2005)

As compositions are vector variables they cannot be analysed component-wise by means of univariate regression ANOVA models and the likeSeemingly unrelated regression models for continuous explanatory variables ormultivariate ANOVA (MANOVA) models for categorical explanatory variablesare appropriate MANOVArsquos multivariate tests and statistics (for example

x2 x3 Transportation

Global partition of total expenditure

x1

Accommodation

amp food

(basic)

Activities

amp shopping

(discretionary)

Partition of

expenses other

than transportation

(at-destination

expenses)

19Determinants in tourist expenditure composition

Pillairsquos trace Hotellingrsquos trace or Wilkrsquos lambda) are invariant to howcomponents are arranged in Figure 1 to compute log-ratios (Mateu-Figueras etal 2013) Univariate tests referring to each particular log-ratio are not invari-ant hence the importance of the interpretability of each log-ratio The softwareSPSS 19 is used to estimate the MANOVA model (GLM procedure)

We have to distinguish between three types of variables moderatingexogenous and endogenous to be included in the analysis

Moderating variables are those which modify the effect of exogenousvariables In our article we treat type of airline as a moderator in other wordswe include its interaction terms with all other variables in the MANOVA model

Exogenous variables are assumed to affect expenditure but not the converseSocio-demographic characteristics (travellerrsquos age gender education incomeetc) are obviously determined prior to any travel decision and hence exogenouswith respect to expenditure

Many of the choices tourists make are interdependent (Dellaert et al 1997)or at least planned simultaneously (Fesenmaier and Jeng 2000) So expenditureis arguably decided on at the same time as many trip attributes At least itis unclear whether expenditure is consistently decided on after trip attributesby all travellers We consider them to be endogenous variables and thereforedo not include them in the MANOVA model as explanatory in order to preventendogeneity problems Instead we display them graphically in the log-ratiospace The means of both log-ratios within each endogenous category are thecategory coordinates

The large sample size (see next subsection) makes it possible to use lowp-values Moderating effects with p-values higher than 001 according to eitherPillairsquos trace Hotellingrsquos trace or Wilksrsquo lambda were removed from the modelAll the variables and moderating effects included in the final model aresignificant at 001

Sample and variables

In this article we use secondary official statistics data The data were providedby the Instituto de Estudios Turiacutesticos (IET) an official agency of the Ministryof Industry Energy and Tourism and one which produces the majority oftourism data in Spain The survey is known as the Encuesta de Gasto Turiacutestico(EGATUR) in which tourism expenditure and other tourist information suchas trip information and tourist sociodemographic characteristics are studiedThe EGATUR survey conducted in 27 major Spanish airports in 2010 usedCAPI (computer assisted personal interview) to interview tourists leaving thecountry The sample is non-proportionally stratified by country of residenceairport and month See IET (2012) for further details on the EGATURmethodology

Our universe is a subset of the EGATUR universe which consists ofEuropean leisure visitors arriving by air and spending at least one night inSpain We excluded flights from outside Europe because LCAs mostly operateshort-haul flights We also centred our study on only those trips with one singledestination thus excluding multi-stage trips as the decision process regardingexpenditure composition for these trips is expected to fundamentally differ fromthat of single-stage trips Stays of over 120 days were also excluded

TOURISM ECONOMICS20

For this study we did not consider tourists

bull who have essential zeros in accommodation (tourists who own a house at thedestination or tourists who stay with friends or relatives)

bull who do not decide how much they will spend on certain components(business and study trips)

bull who do not pay for the trip themselves (trips paid for by employers familyfriends prizes offers etc)

bull for whom composition is wholly or partly unobserved (package tourists)

The final sample size was N= 19359From the expenditure variables included in the EGATUR survey database

and used in the model as budget components we first put the amount paidfor transportation (x1) This component has no zeros

Second the amount of money paid for accommodation and food isundistinguishable for full-board half-board or bed and breakfast accommodationTherefore we define a joint accommodation and food component (basic expendi-ture) In this component we include not only the amount paid for consumptionin bars and restaurants but also for buying groceries and everyday products insupermarkets (x2) This component has no zeros

Finally EGATUR provides an aggregated expenditure on activities andshopping (except groceries and everyday products) To this we added theconceptually similar amount paid for moving around the destination (publictransportation andor car rented at the destination) in order to build anactivities and shopping component (x3) This component had 98 zeros whichstresses its discretionary character Zeros were replaced as explained in thestatistical approach subsection

We consider as at-destination (or also called non-transportation) expenses thebasic component (food and accommodation) and the discretionary component(activities and shopping)

Share and log-ratios are described in Table 1 Explanatory variables are thelevel of education income country of residence gender travel groupprofessional status (for pensioners we consider their last professional position)and age The age variable is built as a combination of the original age variableand the variable referring to the individual economic situation This is donein order to have a specific pensioner category taking into account the varyingretirement ages across countries and professions and thus capturing thosetourists who have more free time to undertake a trip regardless of physical ageTable 2 shows their categories and frequencies

As endogenous variables which are not included in the MANOVA modelbut are included in the log-ratio space we include activities undertaken at thedestination accommodation length of stay total expenditure quartilesquartiles of expenditure made at destination per day (basic plus discretionaryexpenses) and a combination of motivation and destination This last variablehas been constructed as a combination of the motivation and the destinationvariable We distinguish between first those tourists travelling for culturaltourism to singular cities second those who come just for leisure either inthe countryside or more commonly at the seaside and third other typologies(Table 3)

21Determinants in tourist expenditure composition

Table 1 Percentage share and log-ratio descriptive statistics

Min Max Mean SD Skewness Kurtosis

Transportation component (x1) 003 9611 2846 1387 0837 0789Basic component (x2) 021 9963 5472 1613 ndash0114 ndash0450Discretionary component (x3) 001 8668 1681 1292 0988 1174Transportationat-destinationlog-ratio (y1) ndash757 548 01506 10702 0714 1404

Basicdiscretionary log-ratio (y2) ndash530 872 17657 18173 1573 2151

Results

Results for the exogenous variables

Table 4 presents the MANOVA results The univariate R2 corrected for thedegrees of freedom is 0045 (first log-ratio) and 0105 (second log-ratio) Themultivariate uncorrected R2 is 1 ndash Λ = 0178 and corrected for the degrees offreedom is 0176 We checked that the results did not change when removingthe four outliers with the highest Cookrsquos distance

Table 4 presents the estimates for the two log-ratios the log-ratio oftransportation over the other two categories (y1) and the log-ratio of basicexpenditure (accommodation and food) over discretionary expenditure (activitiesand shopping) (y2) respectively Because of the moderating effects within eachlog-ratio there are two columns representing the effects when flying with anLCA or with a legacy airline (main and moderating effects have already addedtogether for easier reading) If the LCA and legacy columns are different thereis a significant moderating effect (p-value lt 001)

As we are interpreting log-ratios a positive estimate of a given predictorcategory means that tourists in that category spend more on the numeratorcompared to the denominator than tourists in the reference predictor categoryA negative estimate shows the opposite

The intercept term shows the main effect of company type that is thepredicted log-ratio for each type of airline within the reference category for allvariables (university education medium income resident in the UK or Ireland25ndash44 yearsrsquo old male travelling with partner and mid-level employee) Withinthe reference categories legacy users spend a higher proportion on transportationcompared to other expenses and LCA users spend more on basic expensescompared to discretionary

The level of education seems to have more to do with activities undertakenthan with transportation Results show that a lower level of education resultsin higher expenditure on basic expenses compared to discretionary and is almostequal for users of both types of airline For mainly or only LCA users a lowerlevel of education results in a higher share in transport compared to the othertwo categories If we put it in another way there seems to be a distinct highlyeducated LCA user segment which utilizes the savings in transport toincrease expenditure in non-transportation expenses and even more so indiscretionary

TOURISM ECONOMICS22

Table 2 Frequency distributions of exogenous and moderating variables in the MANOVAmodel

Variables Count Percentage

Level of educationUp to high school 6019 311Universitya 13340 689

Income categoryMediumlow 797 41Mediuma 13186 681Mediumhigh 4376 226High 1000 52

Country of residencePortugal 624 32Other European countries 850 44Belgium 1006 52Netherlands 1069 55Austria Switzerland and Liechtenstein 1122 58France 1326 68Scandinavian countries 1789 92Germany 2363 122Italy 3113 181United Kingdom and Irelanda 6097 315

GenderFemale 8530 441Malea 10829 559

Travel groupAlone 2484 128In family 2980 154With friends 4440 229With partnera 9446 488

Professional statusHomemaker 517 27Unemployed 602 31Student 1432 74Low-level employee 915 47Mid-level employeea 10964 586High-level employee 2138 110Self-employed 2791 144

AgeOver 45 yearsrsquo old and pensioner 1484 77Over 45 yearsrsquo old and not pensioner 5009 25925ndash44 yearsrsquo olda 10569 54615ndash24 yearsrsquo old 2297 119

Type of airline (moderating variable)Legacy 5884 304Low cost 13475 696

Note aReference categories in the MANOVA model chosen either because they are the largestcategories or because they are the most standard conceptually considered

23Determinants in tourist expenditure composition

Table 3 Frequency distributions of endogenous variables (not included in the MANOVAmodel)

Variables Count Percentage

ActivitiesGolfYes 227 12No 19132 988HikingYes 364 19No 18995 981Sporting eventsYes 603 31No 18756 969Nautical sportsYes 1645 85No 17714 915Other sportsYes 904 47No 18455 953Cultural visitsYes 11180 578No 8179 422Cultural eventsYes 2452 127No 16907 873Other cultural activitiesYes 4175 216No 15184 784SpaYes 1199 62No 18160 938Theme parksYes 1745 90No 17614 910GastronomyYes 1565 81No 17794 919Type of accommodationHotel 4ndash5 5056 261Hotel 3 4922 254Hotel lt 3 6673 345Rented appartment 2116 109Other accommodation 592 31Length of stay5 or less nights 9911 5126ndash8 nights 5952 3079ndash12 nights 1592 8213ndash15 nights 1261 6516 and more nights 643 33

TOURISM ECONOMICS24

Table 3 continued

Variables Count Percentage

Total expenditureQuartile 1 5074 262Quartile 2 5104 264Quartile 3 4986 258Quartile 4 4195 217Expenditure at destination per dayQuartile 1 4855 251Quartile 2 4832 250Quartile 3 4868 251Quartile 4 4804 248Motivation-destinationLeisure on coastin countryside 8083 418Urban tourism 5324 275Others 5952 307

Level of income in contrast affects both log-ratios The higher the incomethe lower the transport share The higher the income the lower the basicdiscretionary expense ratio that is a higher portion of the non-transportexpense is devoted to the discretionary budget

As far as the country of residence is concerned the results show that it affectsmainly the basicdiscretionary log-ratio and exhibits strong airline typemoderating effects LCA users tend to differ depending on their country ofresidence when distributing non-transport expenses into basic versusdiscretionary In general when compared to the UK and Ireland residentcategory (reference category) all other European residents have a largerdiscretionary share More precisely the French Italians and Portuguese have thehighest discretionary share within LCA users and other European countries(mainly central and eastern European) have the highest discretionary sharewithin legacy airline users

Gender has a small effect which is constant for both types of airline Femaletravellers tend to slightly increase the transport share and the share of basicexpenses within the non-transport expenses

The travel group seems to affect both log-ratios and has a moderating effectwith airline type Those tourists who travel in a family group have a highershare in transport expenses when flying with legacy airlines In the second log-ratio travelling as a family with legacy airlines increases the basic expensesshare compared to the share in discretionary Travelling with friends increasesthe share in non-transport expenses for both airline types Besides those whotravel with friends spend more in the discretionary share compared to basicexpenses and those who fly with LCA even more so Finally travelling aloneincreases the basic expenses share compared to the share in discretionaryexpenses for both airline types

When professional status is considered the results show that it mostly affectsthe basicdiscretionary log-ratio and there are some relevant airline type

25Determinants in tourist expenditure composition

Table 4 MANOVA results

y1 log-ratio y2 log-ratiotransportation basicat-destination discretionary

LCA Legacy LCA Legacy

Intercept 0225 0274 2195 1881

Up to high school 0192 0078 0406 0437University 0 0 0 0

Mediumlow income 0073 0289 0037 0518Medium income 0 0 0 0Mediumhigh income ndash0227 ndash0058 ndash0269 ndash0006High income ndash0495 ndash0190 ndash0463 ndash0141

Residence in Portugal 0041 0069 ndash1136 ndash0580Residence in other European Countries 0326 ndash0012 ndash0772 ndash0820Residence in Belgium ndash0127 ndash0159 ndash0667 ndash0461Residence in Netherlands ndash0120 ndash0197 ndash0678 ndash0530Residence in Austria + Switzerland +Liechtenstein ndash0017 ndash0087 ndash0687 ndash0436

Residence in France ndash0112 ndash0111 ndash0870 ndash0553Residence in Scandinavian countries ndash0045 ndash0038 ndash0660 ndash0477Residence in Germany ndash0283 ndash0176 ndash0452 ndash0392Residence in Italy ndash0209 ndash0266 ndash0855 ndash0503Residence in UK or Ireland 0 0 0 0

Female 0094 0094 0093 0093Male 0 0 0 0

Travelling alone ndash0053 ndash0002 0081 0126Travelling in family 0085 0193 0014 0213Travelling with friends ndash0237 ndash0191 ndash0504 ndash0282Travelling with partner 0 0 0 0

Homemaker 0080 0160 0253 0235Unemployed 0030 0159 ndash0036 ndash0039Student ndash0005 ndash0104 ndash0331 ndash0624Low-level employee ndash0034 0194 ndash0194 0063Mid-level employee 0 0 0 0High-level employee ndash0008 ndash0064 ndash0001 ndash0139Self-employed 0029 ndash0086 0193 ndash0072

Over 45 yearsrsquo old and pensioner ndash0060 ndash0060 0633 0633Over 45 yearsrsquo old and not pensioner ndash0010 ndash0010 0142 014225ndash44 yearsrsquo old 0 0 0 015ndash24 yearsrsquo old 0002 0002 ndash0019 ndash0019

moderating effects Legacy users with a low professional status (homemakerunemployed or low-level employee) spend more on transport compared to thenon-transport expenses Conversely students using legacy airlines spend moreon non-transport expenses and these non-transport expenses are to a greater

TOURISM ECONOMICS26

Figure 2 Endogenous variables on the log-ratio space Mean log-ratio valueswithin each category of the endogenous variables

extent devoted to discretionary expenses compared to basic expensesHomemakers increase basic expenditure compared to discretionary with respectto the reference category (mid-level employee) for both airline types LCA userswho are self-employed also increase basic expenditure compared to discretionaryLow-level employees flying with an LCA increase the discretionary sharecompared to the basic and high-level employees flying with legacy airlines alsospend relatively more on the discretionary part

The results show that age affects the second log-ratio more than the firstand there is no moderating effect with the airline type Being a pensioner resultsin a higher share in basic expenses compared to discretionary expenses for bothairline types The remaining categories have hardly any differences from thereference 25ndash44 yearsrsquo old

Results for the endogenous variables

As introduced in the methodology section Figure 2 shows how the endogenousvariables behave in the log-ratio space The origin of the graph is representedby the mean log-ratios in Table 1 01506 for the transportother ratio and17657 for the basicdiscretionary ratio and according to this the graph showscategories with higher or lower ratios than the average

075

060

045

030

015

000

ndash015

ndash030

ndash045

ndash060

Ratio tra

nsport

ationa

t-destination e

xpenses

Ratio basicdiscretionary expenses

050 075 100 125 150 175 200 225 250

27Determinants in tourist expenditure composition

Regarding accommodation it seems that rented apartments and other typesof accommodation behave similarly and have the lowest budget share intransport In the case of hotels as the star rating increases expenditure on basiccompared to discretionary increases and expenditure on transport compared toat-destination decreases

As far as the length of stay is concerned longer stays increase the share ofnon-transport expenditure in basic expenses and reduce the share in transport

As regards destination and motivation tourists going to a seaside or country-side destination for leisure purposes behave quite differently from those goingto an urban destination for urban or cultural tourism The former group spendsmore on transport and less on the discretionary portion However those comingfor urban tourism show an increase in the share of non-transport expenses aswell as in the discretionary share compared to basic expenses

The total expenditure variable gives relevant information It is ordered alongthe first log-ratio axis from top to bottom meaning that those tourists havinga lower total expenditure spend relatively more on the transport share and thosewho have a higher total expenditure spend more on the other two parts of thebudget Regarding the expenditure made at destination per day it is ordereddiagonally from the upper-right corner to the bottom-left corner meaning thattourists with a higher at-destination expenditure per day spend less ontransportation and at the same time more on the discretionary part within at-destination expenses

Finally activities undertaken are located in the lower left quadrant of thegraph meaning that undertaking activities (discretionary expenses) decreasesthe share in transport expenses and although they continue to spend more onbasic expenses than on discretionary the discretionary share increases We havehighlighted some of them Those tourists attending sports and cultural eventsare those spending the most on discretionary compared to basic expensesHiking or some nautical sports could be free of charge thus tourists doing theseactivities have a basic to discretionary ratio close to the mean As for thetransportation share all activities tend to reduce this share about equally withthe exception of golf which leads to a much more substantial reduction

Conclusion and discussion

The main purpose of this article was to study the composition of touristexpenditure and its determinants that is the drivers of the share of touristexpenditure allocated to the different categories of a travel budget and withspecial emphasis on the distinction between legacy versus LCA travellers

The main differences between legacy and LCA users are that income somecountries of residence the travel group and some occupations have differenteffects depending on the type of airline Tourists with a medium or low incomewho travel with legacy airlines tend to spend a greater share of their travelbudget on transportation and basic expenses but those travelling with LCAstend to spend more at the destination and more specifically ondiscretionary expenditure LCA tourists residing in the lsquoother Europeancountriesrsquo spend more on transportation Families flying with legacy airlinesspend more on the transportation share compared to the at-destination share

TOURISM ECONOMICS28

and also spend more on basic expenses compared to discretionary ones Finallyin terms of occupational status legacy users with low-level employment spendmore on transportation whereas the opposite occurs with LCA travellers whospend more on the discretionary component Students and high-level employeesflying with legacy airlines spend more on the discretionary part of the tripbudget

As far as implications for management are concerned assuming that theobjectives of DMOs are to increase the allocation to the at-destinationcomponent of the trip budget (that is lowering the first log-ratio) some of therecommended actions would be to increase marketing efforts directed atmedium- and high-income earners especially at those travelling on LCAs toincrease marketing strategies in some of the European outbound markets suchas Germany (especially those using LCA) and Italy and the Netherlands(especially those flying with legacy airlines) DMOs should also focus on thosetourists travelling with friends (airline type is inconsequential) and studentsusing legacy companies As for the endogenous variables DMOs should directtheir efforts towards those tourists undertaking activities towards those stayinga little bit longer than a week and towards those with high total expenditureFurthermore if DMOs are also interested in capturing the groups who spendrelatively more on discretionary expenses (lowering the second log-ratio) theyshould focus their marketing efforts on those tourists who are highly educatedas well as those who have medium to high incomes and who use LCAsAdditionally they should focus on markets other than the UK and IrelandYoung tourists and tourists who travel with friends (where airline type isinconsequential) low-level employees and students using LCAs as well asstudents and high-level employees flying on legacy airlines should also betargeted In relation to the endogenous variables DMOs must take into accountthose tourists undertaking payable activities those on shorter stays or comingfor urban and cultural tourism or other purposes

The appeal of the CODA methodology for studying tourism budgets lies inthe fact that once the variables have been transformed the researcher can usestandard and well understood statistical models with unbounded error termdistributions while ensuring that the results will be compositionally coherentand that the standard statistical assumptions will hold Conversely results arenot compositionally coherent when we use the MANOVA model on raw sharewithout transforming it into log-ratios We have seen with our data that 84of individuals included in the sample have at least one limit of the 90 shareprediction intervals outside the [01] range

Contrary to the classic and highly restrictive alr and clr transformations theCODA methodology offers the potential to construct tailor-made log-ratios whichare intuitive to interpret and suit the research questions at hand A classificationtree of components is a clear and useful tool in this respect We encourageresearchers to use this approach in further research on tourism expenditure whenbudget share division is fundamental to the researchersrsquo questions

In this study we have encountered two limitations First contrary to studiesthat use macro data in micro data studies the coexistence of full-board halfboard bed and breakfast and accommodation-only tourists makes it impossibleto separate accommodation and food expenditure meaningfully Othersubdivisions (for example discretionary expenditure on activities non-grocery

29Determinants in tourist expenditure composition

shopping and moving around the destination) are technically feasible but canonly increase the percentages of zeros The CODA methodology involves somedegree of amalgamation of components Second even though there are someadvantages to using a database from an official statistics institution as in thiscase (large sample size and scope) the main disadvantage is that the set ofavailable variables cannot be controlled by the researcher A further issue to beaddressed is the time dimension As the EGATUR survey is conducted annuallyfurther research can be done to include a repeat cross-section analysis in orderto capture trends in the effects of predictors

References

Aitchison J (1986) The Statistical Analysis of Compositional Data Monographs on Statistics and AppliedProbability Chapman and Hall London (Reprinted in 2003 with additional material by TheBlackburn Press)

Alegre J Mateo S and Pou LL (2009) lsquoParticipation in tourism consumption and the intensityof participation an analysis of their socio-demographic and economic determinantsrsquo TourismEconomics Vol 15 No 3 pp 531ndash546

Bewley R (1982) lsquoThe generalized addilog demand system applied to Australian time series andcross section datarsquo Australian Economic Papers Vol 21 No 38 pp 177ndash192

Bewley R and Fiebig DG (1988) lsquoEstimation of price elasticities for an international telephonedemand modelrsquo The Journal of Industrial Economics Vol 36 No 4 pp 393ndash409

Cai LA (1998) lsquoAnalysing household food expenditure patterns on trips and vacations a Tobitmodersquo Journal of Hospitality and Tourism Research Vol 22 No 4 pp 338ndash358

Cai LA (1999) lsquoRelationship of household characteristics and lodging expenditure on leisure tripsrsquoJournal of Hospitality and Leisure Marketing Vol 6 No 2 pp 5ndash18

Cai LA Hong G-S and Morrison AM (1995) lsquoHousehold expenditure patterns for tourismproducts and servicesrsquo Journal of Travel and Tourism Marketing Vol 4 No 4 pp 15ndash39

Cannon TF and Ford J (2002) lsquoRelationship of demographic and trip characteristics to visitorspending an analysis of sports travel visitors across timersquo Tourism Economics Vol 8 No 3 pp 263ndash271

Castillo-Manzano JL and Marchena-Goacutemez M (2010) lsquoAnalysis of determinants of airline choiceprofiling the LCC passengerrsquo Applied Economics Letters Vol 18 No 1 pp 49ndash53

Chiou Y-C and Chen Y-H (2010) lsquoFactors influencing the intentions of passengers regarding full ser-vice and low cost carriers a notersquo Journal of Air Transport Management Vol 16 No 4 pp 226ndash228

Coenen M and van Eekeren L (2003) lsquoA study of the demand for domestic tourism by Swedishhouseholds using a two-staged budgeting modelrsquo Scandinavian Journal of Hospitality and TourismVol 3 No 2 pp 114ndash133

Craggs R and Schofield P (2009) lsquoExpenditure-based segmentation and visitor profiling at theQuays in Salford UKrsquo Tourism Economics Vol 15 No 1 pp 243ndash260

Deaton A and Muellbauer J (1980) lsquoAn almost ideal demand systemrsquo American Economic ReviewVol 70 No 3 pp 312ndash326

Dellaert B Borgers A and Timmermans H (1997) lsquoConsumer activity pattern choice devel-opment and test of stage-dependent conjoint choice experimentsrsquo Journal of Retailing and ConsumerServices Vol 4 No 1 pp 25ndash37

Divisekera S (2007) Modelling and Estimation of Tourism Demand Elasticities A Study of TourismExpenditure Allocation in Australia Cooperative Research Centre for Sustainable Tourism (CRCST)Gold Coast

Divisekera S (2009) lsquoEx-post demand for Australian tourism goods and servicesrsquo Tourism EconomicsVol 15 No 1 pp 153ndash180

Divisekera S (2010) lsquoEconomics of leisure and non-leisure tourist demand a study of domesticdemand for Australian tourismrsquo Tourism Economics Vol 16 No 1 pp 117ndash136

Dolnicar S Crouch GI Devinney T Huybers T Louviere JJ and Oppewal H (2008)lsquoTourism and discretionary income allocation Heterogeneity among householdsrsquo TourismManagement Vol 29 No 1 pp 44ndash52

Downward P and Lumsdon L (2000) lsquoThe demand for day-visits an analysis of visitor spendingrsquoTourism Economics Vol 6 No 3 pp 251ndash261

TOURISM ECONOMICS12

one component (budget share) can increase only if the relative importance ofat least one other decreases Most methods used to model budget share suchas the almost ideal demand system (Deaton and Muellbauer 1980) ignore atleast partly the constraints and distributional nature of compositional data Tothe best of our knowledge there is no scholarly study of the composition oftourist budget using an appropriate methodology for compositional data analysis

This article is structured as follows First we present a review of the mainapproaches in modelling tourist expenditure which is then followed by adescription of the CODA methodology Then we introduce the method anddata which are followed by the results and finally by the overall conclusionsand discussion

Major approaches in modelling tourism expenditure

The study of expenditure in terms of composition (relative share of each partof the budget) is not the same as the study of expenditure in absolute termsand the variables affecting absolute expenditure may differ from those affectingrelative expenditure Tourism budgets have been approached from bothperspectives in the literature More precisely tourism expenditure has beenanalysed globally in absolute terms per budget parts in relative terms (share)per budget parts and as a part in itself in family budgets

The vast majority of microeconomic tourism demand studies (24 out of 27in the review of Wang and Davidson (2010)) concern the prediction of onesingle aggregated expenditure variable Methods range from mean comparisontests (Craggs and Schofield 2009) OLS and WLS regression (Downward andLumsdon 2000 2003 Cannon and Ford 2002) to advanced econometrictechniques For instance Hung et al (2012) use quantile regression to buildequations predicting not only typical expenditures but also the highest andlowest Alegre et al (2009) Eugenio-Martin (2003) Hong et al (1999) andNicolau and Maacutes (2005) propose double-hurdle Heckit and related models toseparate the decision whether to spend on tourism from the decision of howmuch to spend With respect to the explanatory variables used the mostcommon are income age gender marital status education place of residencelength of stay travel group size and composition accommodation main trippurpose and activities (Marcussen 2011) In general explanatory variables canbe grouped into economic variables (prices and income) socio-demographicvariables and trip or travel-related variables (Wang et al 2006 Sainaghi 2012)Nicolau (2009) includes individual price sensitivity estimated in a previousmodel

Another stream of research is that which analyses tourist expenditure pertourism product (for example lodging food transportation and sightseeingentertainment) A common argument for studying tourist expenditure patternsper tourism product is that it provides vital information to travel organizersand destination marketers when designing the appropriate marketing strategiesFor this purpose researchers have used several methods such as Tobit MANOVAor seemingly unrelated linear regressions (Pyo et al 1991 Cai et al 1995Oppermann 1996 Cai 1998 1999 Lee 2001 Lehto et al 2001 Jang et al2004 Wang et al 2006) Socio-demographic variables are the most common

13Determinants in tourist expenditure composition

significant variables Age affects expenditure on meals as does travel groupwhich also affects transportation expenditure Marital status usually affects foodand accommodation expenditure

Since part expenditure in absolute terms is related to total expenditure acommon finding in these studies is that some of the explanatory variables affectall budget elements roughly equally For example Cai et al (1995) found thatthe higher the level of education the higher the expenditure in all budget partsand Wang et al (2006) conclude similarly about household income Such resultswill never be obtained when analysing budget share

The empirical analysis of budget share both for tourism expenditure andgeneral family budgets commonly involves estimating an almost ideal demandsystem of equations (Deaton and Muellbauer 1980) The almost ideal demandsystem is designed to analyse the interdependences of budget allocations thusovercoming the limitations of single-equation modelling and has receivedmuch attention in the last decade (Song et al 2012) It directly fits shares(composition) as dependent variables in a set of simultaneous regressions Theapproach has focused mainly on estimating price and income elasticities Studiesthat use macroeconomic data include among others those by Divisekera (20072009 2010) Fujii et al (1985) OrsquoHagan and Harrison (1984) Syriopoulos andSinclair (1993) and Wu et al (2011)

The almost ideal demand system has been applied both to data from a givenorigin to multiple destinations (that is ex ante ndash see Li et al 2004 Divisekera2009) or from multiple origins to a given destination which is the case in ourarticle (ex post ndash see Divisekera 2009) In ex post studies it is commonly assumedthat various commodities can be aggregated to broad bundles of productsprovided that prices in a bundle move in parallel and that the utility functionwith respect to tourism and other goods is weakly separable This makes itpossible to conceptualize tourism consumption as a multi-stage process In thefirst stage tourists allocate a household budget part to tourism consumptionin the second a tourism budget part to each tripdestination in the third adestination budget part to each good and service including transportationaccommodation and so on Ex post studies such as ours model only the laststage There are however studies of how tourism competes against othercategories of discretionary expenditure using individual micro data Forexample Melenberg and Van Soest (1996) use different parametric and semi-parametric Tobit models to explain the vacation budget share from householdcharacteristics and Dolnicar et al (2008) analyse how households allocatediscretionary income between tourism and competing uses

The review of Wang and Davidson (2010) concludes that since the vastmajority of tourism demand studies are conducted at macro level (this holdseven more for budget share analysis) there is room for more micro-econometricstudies in this area as the only manner of accounting for demand heterogeneityThe almost ideal demand system approach when applied to micro data caninclude individual characteristics For example Coenen and van Eekeren (2003)and Fleischer et al (2011) make ex ante studies of individual budget share inSweden and Israel respectively and use previous Heckit-type selectionequations to model the decision whether to travel or not Coenen and vanEekeren (2003) include household size and income as individual characteristicsFleischer et al (2011) add age education real state ownership place of birth

TOURISM ECONOMICS14

and Internet use The basic aim of those articles is to estimate elasticities andthey use the individual characteristics only as controls Fleischer et al (2011)are the only researchers to provide the estimates of equations predicting budgetshare from traveller characteristics as a by-product of their elasticity estimatesBeing born in the country of origin is reported to increase the transportationshare and reduce the share of on-site expenditures Education and householdreal estate assets reduce the share of on-site expenditures

The aim and approach of this article are similar to those of Coenen and vanEekeren (2003) and of Fleischer et al (2011) regarding the use of individualcharacteristics to predict budget share but differ in three important respects

bull Our study is ex post so that Heckit modelling is unfeasiblebull The effect of individual characteristics on budget composition (share) is the

core of the analysis Explanatory variables are individual characteristics ratherthan prices The results will include the effects of traveller heterogeneityrather than demand equations

bull Our analysis takes into account the compositional restrictions of the databy using the CODA methodology

The CODA methodology

Compared to absolute data compositional data such as budget share lie in aconstrained space A D-term composition measured on individual i xi1 xi2hellipxiD

has the following constraints

0 le xid le 1 and DΣ

d=1xid = 1 (1)

Aitchison (1986) and Pawlowsky-Glahn and Buccianti (2011) warn against theserious problems that arise when using standard statistical analysis tools oncompositional data Compositional data are non-normal and heteroscedasticOne component can increase only if some other(s) decreases This results innegative spurious correlations among the components and prevents interpretingeffects of linear models in the usual way lsquokeeping everything else constantrsquo

Even if specialized CODA techniques are starting to appear (for exampleRonning 1992 Thioacute-Henestrosa and Martiacuten-Fernaacutendez 2005) the easy way(Aitchison 1986 McLaren et al 1995 Fry et al 1996) involves transformingcompositional data so that they can be subject to standard and well-understoodstatistical techniques This is the approach we take in this article In short thisinvolves using the transformed share by means of logarithms of ratios insteadof the raw share

To ensure compositional coherence of predicted budget share (unit sum andnon-negativity) a set of parameter constraints is imposed to the almost idealdemand system However the presence of an error term with an unboundeddistribution (usually normal) results in a non-zero probability that actual sharelies outside the [01] interval (McLaren et al 1995 Fry et al 1996 Fry 2011)In other words the in fact bounded distribution of budget share results in amisspecification of the almost ideal demand system and of any model fittingpercentage share with an unbounded error distribution

15Determinants in tourist expenditure composition

The fact that the error term in any proper system of demand equationsapplied to budget share should take into account the data compositional naturehas been widely acknowledged (Aitchison 1986 Ronning 1992 Fry et al1996) However the review in Fry (2011) reports few studies that have appliedthe CODA methodology to demand equations and to the study of householdbudgets (McLaren et al 1995 Fry et al 1996 2000)

When fitting demand equations to log-ratios of expenditure componentsEngelrsquos aggregation condition which requires that household total expenditureshould equal the sum of components is automatically satisfied The model caneasily be extended to a full consumer demand analysis by the introduction ofprices and other covariates such as consumer characteristics (Aitchison 1986)McLaren et al (1995) relate the CODA analysis with log-ratio transformationto the almost ideal demand system and conclude that CODA makes it possibleto reach the same objectives with normal and homoscedastic error terms

Related developments are the indirect addilog system (Houthakker 1960)and the generalized addilog system (Bewley 1982) When applied tocompositions (for example Bewley and Fiebig 1988) the latter is equivalentto the CODA methodology in which the log-ratios of each component over thegeometric means of all components are the dependent variables in the set ofsimultaneous regressions

To the best of our knowledge there is no study of tourism budget share usingthe CODA methodology or any other methodology accounting for thecompositional constraints in compositional data

Method and data

Statistical approach

The simplest CODA approach involves applying standard statistical techniqueson logarithms of ratios of components Several log-ratio transformations havebeen suggested in the early CODA literature (Egozcue et al 2003) The additivelog-ratio transformation (alr) used by Fry et al (1996 2000) is the most popularand the easiest to compute given that it is simply the log-ratio of eachcomponent to the last

yid = ln(xidxiD) = ln(xid) ndash ln(xiD) with d = 123D ndash 1 (2)

The centred log-ratio transformation (clr) used by the generalized addilogdemand system (Bewley 1982) computes the log-ratios of each component overthe geometric mean of all the components including itself

⎛ xid ⎞yid = ln ⎜ ndashndashndashndashndashndashndashndashndashndashndashndash ⎟ with d = 123D (3)

⎝ D xi1xi2xi3 xiD ⎠

All log-ratio transformed yid variables recover the full unconstrained ndashinfin to infinrange It must be noted that one dimension is lost in the alr while in the clrone dimension is a linear combination of the remaining

TOURISM ECONOMICS16

The alr is commonly used for statistical modelling and prediction ofcompositions (Fry et al 1996) Conversely the clr transformation is commonlyused for statistical techniques which are based on a metric such as a clusteranalysis because of its preservation of distances even though it leads to asingular covariance matrix Thus while the alr would be appropriate for thepurpose of this article the fact that one component must be used as referencefor all others reduces its flexibility and interpretability Alternatives arepresented at the end of this subsection

While having zero expenditures in absolute data indeed has significantmethodological consequences (Lee 2001) these consequences are arguably moreserious in the CODA methodology If the xid variables contain zeros then log-ratios cannot be computed An obvious initial procedure to reduce zeros is toamalgamate small and conceptually similar components with many zeros intolarger ones In tourism budget research it can be useful to group together allexpenditure on activities or all expenditure on food for instance

In certain instances some zero components result from individual character-istics which are called essential zeros in the CODA literature (Aitchison 1986)Another typology of zeros encountered in the CODA literature is the roundingzero that is a component which is present but is too small to be detected bythe measurement instrument This is typical in chemical biological andgeological compositions and the CODA literature offers ample instruments todeal with rounding zeros

A classic essential zero example in economics is in household budget researchwhen measuring expenditure on tobacco and will essentially be zero if allmembers are non-smokers If tobacco expenditure is decidedly in the researchersrsquointerest the target population should be redefined to include only smokers Inmany instances budget research is often not clear whether zero expenditurescome closer to being essential or rounding zeros In some cases they can beunderstood as corner solutions in a utility maximization problem In others theycan be understood as the inherent randomness of human behaviour or as thelimitations of the data Tourists may spend a certain amount on activities oncertain trips but not on others and so surveys of only one trip will unavoidablycontain some zeros of this type Tourists may also forget or fail to report trivialexpenses such as postcard shopping local bus tickets and going to a museum(see Legoheacuterel (1998) for a discussion on the instability of tourism expenditure)Fry et al (2000) claim that in both situations zeros can be proxied by a verysmall value and thus be treated as rounding zeros In tourist budget researchto treat zero expenses in activities as rounding zeros implies assuming firstthat there is basically no tourist type who will never spend anything onactivities and second that tourists who generally spend little on activities arebasically similar to those who spend nothing or fail to report small expensesWe find both assumptions to be reasonable

Fry et al (2000) essentially used the same zero replacement strategy that waslater suggested by Martiacuten-Fernaacutendez et al (2003) namely replacing xid = 0 with

xprimeid = kδid with 0 lt k lt 1 (4)

where δid is the smallest detectable proportion for individual i and component d

17Determinants in tourist expenditure composition

Martiacuten-Fernaacutendez et al (2003) suggest using k = 065 although a sensitivityanalysis of the results on the choice of k is always advisable (we used k = 030and k = 099 with no sizeable change in the estimates) Next non-zero xid valueshave to be reduced in order to preserve the unit sum and the ratios among non-zero components As suggested by Martiacuten-Fernaacutendez et al (2003) with

xprimeid = xid (1 ndash Σxid=0

xprimeid) (5)

Simulations show this method performs particularly well if the proportion ofzeros is below 10 (Martiacuten-Fernaacutendez et al 2011)

In our data set zeros were present only in one budget category (discretionaryexpenditure) The minimum amount spent by the non-zero group was euro1which roughly corresponds to the price of a city bus ticket the entrance to asubsidized local museum or a cheap souvenir Since the total expenditureis known for each individual we compute δid by dividing euro1 with the totalexpenditure of individual I (see the appendix for the SPSS commandsyntax)

In this article we consider alternative log-ratio transformations which aremore flexible than the alr and clr in that the denominator does not have tobe the same in all ratios This increased flexibility makes it easier to computelog-ratios which are more interpretable with respect to the researchersrsquo questions orhypotheses

In general an interpretable log-ratio transformation is easy to computewhenever there is an interpretable sequential binary partition of componentsinto pairs of groups of components according to the researchersrsquo objectives orto the conceptual similarity of the components These partitions start bydividing components into two clusters and then continue by subdividingone of the clusters into two until each component constitutes its owncluster D components always involve Dndash1 partitions These partitions are bestunderstood as a partition tree or dendrogram (Pawlowsky-Glahn and Egozcue2011)

A meaningful log-ratio transformation takes ratios of the geometric meansof the two component clusters at each partition Numerators and denominatorsare interchangeable In our article we consider xi1 = transportation expenditurexi2 = accommodation and food (basic expenditure) and xi3 = activities andshopping (discretionary expenditure) An interpretable sequential partition isshown in Figure 1

The sequential partition in Figure 1 can have a dual interpretation It mayimply that researchers consider both types of at-destination expenses to bemutually similar and less similar to transportation Or it may imply that theresearch questions involve the distribution of total expenditure betweentransportation and at-destination expenditure and the distribution ofat-destination expenditure into basic and discretionary components as is thecase in our article

The first log-ratio compares transportation expenditure with the geometricmean of accommodation and food (basic expenditure) and activities andshopping (discretionary expenditure) With this ratio we want to observe howtravelling with an LCA or a legacy company affects the share of transportationcompared to non-transportation (at-destination) expenses

TOURISM ECONOMICS18

Figure 1 Sequential partition of total expenditure

⎛ xi1 ⎞ 1 1yi1 = ln ⎜ ndashndashndashndash ⎟ = ln(xi1) ndash ndash ln(xi2) ndash ndash ln(xi3) (6)

⎝ xi2xi3 ⎠ 2 2

Positive values show a transportation share greater than the geometric meanof the remaining two components Negative values show the opposite

The second log-ratio is a ratio of accommodation and food (basic expenditure)over activities and shopping (discretionary expenditure) With this ratio wewant to find out to what tourists allocate the rest of their trip budget andwhether more is allocated to basic or discretionary expenditure once they havepaid for (often in advance) their transportation

⎛ xi2 ⎞yi2 = ln ⎜ ndashndashndash ⎟ = ln(xi2) ndash ln(xi3) (7)

⎝ xi3 ⎠

Positive values show a basic (accommodation and food) share which is largerthan the discretionary (activities and shopping) share Negative values illustratethe opposite

Log-ratio transformations based on sequential binary partitions areproportional to the isometric log-ratio transformation (ilr see Egozcue et al2003 Egozcue and Pawlowsky-Glahn 2005)

As compositions are vector variables they cannot be analysed component-wise by means of univariate regression ANOVA models and the likeSeemingly unrelated regression models for continuous explanatory variables ormultivariate ANOVA (MANOVA) models for categorical explanatory variablesare appropriate MANOVArsquos multivariate tests and statistics (for example

x2 x3 Transportation

Global partition of total expenditure

x1

Accommodation

amp food

(basic)

Activities

amp shopping

(discretionary)

Partition of

expenses other

than transportation

(at-destination

expenses)

19Determinants in tourist expenditure composition

Pillairsquos trace Hotellingrsquos trace or Wilkrsquos lambda) are invariant to howcomponents are arranged in Figure 1 to compute log-ratios (Mateu-Figueras etal 2013) Univariate tests referring to each particular log-ratio are not invari-ant hence the importance of the interpretability of each log-ratio The softwareSPSS 19 is used to estimate the MANOVA model (GLM procedure)

We have to distinguish between three types of variables moderatingexogenous and endogenous to be included in the analysis

Moderating variables are those which modify the effect of exogenousvariables In our article we treat type of airline as a moderator in other wordswe include its interaction terms with all other variables in the MANOVA model

Exogenous variables are assumed to affect expenditure but not the converseSocio-demographic characteristics (travellerrsquos age gender education incomeetc) are obviously determined prior to any travel decision and hence exogenouswith respect to expenditure

Many of the choices tourists make are interdependent (Dellaert et al 1997)or at least planned simultaneously (Fesenmaier and Jeng 2000) So expenditureis arguably decided on at the same time as many trip attributes At least itis unclear whether expenditure is consistently decided on after trip attributesby all travellers We consider them to be endogenous variables and thereforedo not include them in the MANOVA model as explanatory in order to preventendogeneity problems Instead we display them graphically in the log-ratiospace The means of both log-ratios within each endogenous category are thecategory coordinates

The large sample size (see next subsection) makes it possible to use lowp-values Moderating effects with p-values higher than 001 according to eitherPillairsquos trace Hotellingrsquos trace or Wilksrsquo lambda were removed from the modelAll the variables and moderating effects included in the final model aresignificant at 001

Sample and variables

In this article we use secondary official statistics data The data were providedby the Instituto de Estudios Turiacutesticos (IET) an official agency of the Ministryof Industry Energy and Tourism and one which produces the majority oftourism data in Spain The survey is known as the Encuesta de Gasto Turiacutestico(EGATUR) in which tourism expenditure and other tourist information suchas trip information and tourist sociodemographic characteristics are studiedThe EGATUR survey conducted in 27 major Spanish airports in 2010 usedCAPI (computer assisted personal interview) to interview tourists leaving thecountry The sample is non-proportionally stratified by country of residenceairport and month See IET (2012) for further details on the EGATURmethodology

Our universe is a subset of the EGATUR universe which consists ofEuropean leisure visitors arriving by air and spending at least one night inSpain We excluded flights from outside Europe because LCAs mostly operateshort-haul flights We also centred our study on only those trips with one singledestination thus excluding multi-stage trips as the decision process regardingexpenditure composition for these trips is expected to fundamentally differ fromthat of single-stage trips Stays of over 120 days were also excluded

TOURISM ECONOMICS20

For this study we did not consider tourists

bull who have essential zeros in accommodation (tourists who own a house at thedestination or tourists who stay with friends or relatives)

bull who do not decide how much they will spend on certain components(business and study trips)

bull who do not pay for the trip themselves (trips paid for by employers familyfriends prizes offers etc)

bull for whom composition is wholly or partly unobserved (package tourists)

The final sample size was N= 19359From the expenditure variables included in the EGATUR survey database

and used in the model as budget components we first put the amount paidfor transportation (x1) This component has no zeros

Second the amount of money paid for accommodation and food isundistinguishable for full-board half-board or bed and breakfast accommodationTherefore we define a joint accommodation and food component (basic expendi-ture) In this component we include not only the amount paid for consumptionin bars and restaurants but also for buying groceries and everyday products insupermarkets (x2) This component has no zeros

Finally EGATUR provides an aggregated expenditure on activities andshopping (except groceries and everyday products) To this we added theconceptually similar amount paid for moving around the destination (publictransportation andor car rented at the destination) in order to build anactivities and shopping component (x3) This component had 98 zeros whichstresses its discretionary character Zeros were replaced as explained in thestatistical approach subsection

We consider as at-destination (or also called non-transportation) expenses thebasic component (food and accommodation) and the discretionary component(activities and shopping)

Share and log-ratios are described in Table 1 Explanatory variables are thelevel of education income country of residence gender travel groupprofessional status (for pensioners we consider their last professional position)and age The age variable is built as a combination of the original age variableand the variable referring to the individual economic situation This is donein order to have a specific pensioner category taking into account the varyingretirement ages across countries and professions and thus capturing thosetourists who have more free time to undertake a trip regardless of physical ageTable 2 shows their categories and frequencies

As endogenous variables which are not included in the MANOVA modelbut are included in the log-ratio space we include activities undertaken at thedestination accommodation length of stay total expenditure quartilesquartiles of expenditure made at destination per day (basic plus discretionaryexpenses) and a combination of motivation and destination This last variablehas been constructed as a combination of the motivation and the destinationvariable We distinguish between first those tourists travelling for culturaltourism to singular cities second those who come just for leisure either inthe countryside or more commonly at the seaside and third other typologies(Table 3)

21Determinants in tourist expenditure composition

Table 1 Percentage share and log-ratio descriptive statistics

Min Max Mean SD Skewness Kurtosis

Transportation component (x1) 003 9611 2846 1387 0837 0789Basic component (x2) 021 9963 5472 1613 ndash0114 ndash0450Discretionary component (x3) 001 8668 1681 1292 0988 1174Transportationat-destinationlog-ratio (y1) ndash757 548 01506 10702 0714 1404

Basicdiscretionary log-ratio (y2) ndash530 872 17657 18173 1573 2151

Results

Results for the exogenous variables

Table 4 presents the MANOVA results The univariate R2 corrected for thedegrees of freedom is 0045 (first log-ratio) and 0105 (second log-ratio) Themultivariate uncorrected R2 is 1 ndash Λ = 0178 and corrected for the degrees offreedom is 0176 We checked that the results did not change when removingthe four outliers with the highest Cookrsquos distance

Table 4 presents the estimates for the two log-ratios the log-ratio oftransportation over the other two categories (y1) and the log-ratio of basicexpenditure (accommodation and food) over discretionary expenditure (activitiesand shopping) (y2) respectively Because of the moderating effects within eachlog-ratio there are two columns representing the effects when flying with anLCA or with a legacy airline (main and moderating effects have already addedtogether for easier reading) If the LCA and legacy columns are different thereis a significant moderating effect (p-value lt 001)

As we are interpreting log-ratios a positive estimate of a given predictorcategory means that tourists in that category spend more on the numeratorcompared to the denominator than tourists in the reference predictor categoryA negative estimate shows the opposite

The intercept term shows the main effect of company type that is thepredicted log-ratio for each type of airline within the reference category for allvariables (university education medium income resident in the UK or Ireland25ndash44 yearsrsquo old male travelling with partner and mid-level employee) Withinthe reference categories legacy users spend a higher proportion on transportationcompared to other expenses and LCA users spend more on basic expensescompared to discretionary

The level of education seems to have more to do with activities undertakenthan with transportation Results show that a lower level of education resultsin higher expenditure on basic expenses compared to discretionary and is almostequal for users of both types of airline For mainly or only LCA users a lowerlevel of education results in a higher share in transport compared to the othertwo categories If we put it in another way there seems to be a distinct highlyeducated LCA user segment which utilizes the savings in transport toincrease expenditure in non-transportation expenses and even more so indiscretionary

TOURISM ECONOMICS22

Table 2 Frequency distributions of exogenous and moderating variables in the MANOVAmodel

Variables Count Percentage

Level of educationUp to high school 6019 311Universitya 13340 689

Income categoryMediumlow 797 41Mediuma 13186 681Mediumhigh 4376 226High 1000 52

Country of residencePortugal 624 32Other European countries 850 44Belgium 1006 52Netherlands 1069 55Austria Switzerland and Liechtenstein 1122 58France 1326 68Scandinavian countries 1789 92Germany 2363 122Italy 3113 181United Kingdom and Irelanda 6097 315

GenderFemale 8530 441Malea 10829 559

Travel groupAlone 2484 128In family 2980 154With friends 4440 229With partnera 9446 488

Professional statusHomemaker 517 27Unemployed 602 31Student 1432 74Low-level employee 915 47Mid-level employeea 10964 586High-level employee 2138 110Self-employed 2791 144

AgeOver 45 yearsrsquo old and pensioner 1484 77Over 45 yearsrsquo old and not pensioner 5009 25925ndash44 yearsrsquo olda 10569 54615ndash24 yearsrsquo old 2297 119

Type of airline (moderating variable)Legacy 5884 304Low cost 13475 696

Note aReference categories in the MANOVA model chosen either because they are the largestcategories or because they are the most standard conceptually considered

23Determinants in tourist expenditure composition

Table 3 Frequency distributions of endogenous variables (not included in the MANOVAmodel)

Variables Count Percentage

ActivitiesGolfYes 227 12No 19132 988HikingYes 364 19No 18995 981Sporting eventsYes 603 31No 18756 969Nautical sportsYes 1645 85No 17714 915Other sportsYes 904 47No 18455 953Cultural visitsYes 11180 578No 8179 422Cultural eventsYes 2452 127No 16907 873Other cultural activitiesYes 4175 216No 15184 784SpaYes 1199 62No 18160 938Theme parksYes 1745 90No 17614 910GastronomyYes 1565 81No 17794 919Type of accommodationHotel 4ndash5 5056 261Hotel 3 4922 254Hotel lt 3 6673 345Rented appartment 2116 109Other accommodation 592 31Length of stay5 or less nights 9911 5126ndash8 nights 5952 3079ndash12 nights 1592 8213ndash15 nights 1261 6516 and more nights 643 33

TOURISM ECONOMICS24

Table 3 continued

Variables Count Percentage

Total expenditureQuartile 1 5074 262Quartile 2 5104 264Quartile 3 4986 258Quartile 4 4195 217Expenditure at destination per dayQuartile 1 4855 251Quartile 2 4832 250Quartile 3 4868 251Quartile 4 4804 248Motivation-destinationLeisure on coastin countryside 8083 418Urban tourism 5324 275Others 5952 307

Level of income in contrast affects both log-ratios The higher the incomethe lower the transport share The higher the income the lower the basicdiscretionary expense ratio that is a higher portion of the non-transportexpense is devoted to the discretionary budget

As far as the country of residence is concerned the results show that it affectsmainly the basicdiscretionary log-ratio and exhibits strong airline typemoderating effects LCA users tend to differ depending on their country ofresidence when distributing non-transport expenses into basic versusdiscretionary In general when compared to the UK and Ireland residentcategory (reference category) all other European residents have a largerdiscretionary share More precisely the French Italians and Portuguese have thehighest discretionary share within LCA users and other European countries(mainly central and eastern European) have the highest discretionary sharewithin legacy airline users

Gender has a small effect which is constant for both types of airline Femaletravellers tend to slightly increase the transport share and the share of basicexpenses within the non-transport expenses

The travel group seems to affect both log-ratios and has a moderating effectwith airline type Those tourists who travel in a family group have a highershare in transport expenses when flying with legacy airlines In the second log-ratio travelling as a family with legacy airlines increases the basic expensesshare compared to the share in discretionary Travelling with friends increasesthe share in non-transport expenses for both airline types Besides those whotravel with friends spend more in the discretionary share compared to basicexpenses and those who fly with LCA even more so Finally travelling aloneincreases the basic expenses share compared to the share in discretionaryexpenses for both airline types

When professional status is considered the results show that it mostly affectsthe basicdiscretionary log-ratio and there are some relevant airline type

25Determinants in tourist expenditure composition

Table 4 MANOVA results

y1 log-ratio y2 log-ratiotransportation basicat-destination discretionary

LCA Legacy LCA Legacy

Intercept 0225 0274 2195 1881

Up to high school 0192 0078 0406 0437University 0 0 0 0

Mediumlow income 0073 0289 0037 0518Medium income 0 0 0 0Mediumhigh income ndash0227 ndash0058 ndash0269 ndash0006High income ndash0495 ndash0190 ndash0463 ndash0141

Residence in Portugal 0041 0069 ndash1136 ndash0580Residence in other European Countries 0326 ndash0012 ndash0772 ndash0820Residence in Belgium ndash0127 ndash0159 ndash0667 ndash0461Residence in Netherlands ndash0120 ndash0197 ndash0678 ndash0530Residence in Austria + Switzerland +Liechtenstein ndash0017 ndash0087 ndash0687 ndash0436

Residence in France ndash0112 ndash0111 ndash0870 ndash0553Residence in Scandinavian countries ndash0045 ndash0038 ndash0660 ndash0477Residence in Germany ndash0283 ndash0176 ndash0452 ndash0392Residence in Italy ndash0209 ndash0266 ndash0855 ndash0503Residence in UK or Ireland 0 0 0 0

Female 0094 0094 0093 0093Male 0 0 0 0

Travelling alone ndash0053 ndash0002 0081 0126Travelling in family 0085 0193 0014 0213Travelling with friends ndash0237 ndash0191 ndash0504 ndash0282Travelling with partner 0 0 0 0

Homemaker 0080 0160 0253 0235Unemployed 0030 0159 ndash0036 ndash0039Student ndash0005 ndash0104 ndash0331 ndash0624Low-level employee ndash0034 0194 ndash0194 0063Mid-level employee 0 0 0 0High-level employee ndash0008 ndash0064 ndash0001 ndash0139Self-employed 0029 ndash0086 0193 ndash0072

Over 45 yearsrsquo old and pensioner ndash0060 ndash0060 0633 0633Over 45 yearsrsquo old and not pensioner ndash0010 ndash0010 0142 014225ndash44 yearsrsquo old 0 0 0 015ndash24 yearsrsquo old 0002 0002 ndash0019 ndash0019

moderating effects Legacy users with a low professional status (homemakerunemployed or low-level employee) spend more on transport compared to thenon-transport expenses Conversely students using legacy airlines spend moreon non-transport expenses and these non-transport expenses are to a greater

TOURISM ECONOMICS26

Figure 2 Endogenous variables on the log-ratio space Mean log-ratio valueswithin each category of the endogenous variables

extent devoted to discretionary expenses compared to basic expensesHomemakers increase basic expenditure compared to discretionary with respectto the reference category (mid-level employee) for both airline types LCA userswho are self-employed also increase basic expenditure compared to discretionaryLow-level employees flying with an LCA increase the discretionary sharecompared to the basic and high-level employees flying with legacy airlines alsospend relatively more on the discretionary part

The results show that age affects the second log-ratio more than the firstand there is no moderating effect with the airline type Being a pensioner resultsin a higher share in basic expenses compared to discretionary expenses for bothairline types The remaining categories have hardly any differences from thereference 25ndash44 yearsrsquo old

Results for the endogenous variables

As introduced in the methodology section Figure 2 shows how the endogenousvariables behave in the log-ratio space The origin of the graph is representedby the mean log-ratios in Table 1 01506 for the transportother ratio and17657 for the basicdiscretionary ratio and according to this the graph showscategories with higher or lower ratios than the average

075

060

045

030

015

000

ndash015

ndash030

ndash045

ndash060

Ratio tra

nsport

ationa

t-destination e

xpenses

Ratio basicdiscretionary expenses

050 075 100 125 150 175 200 225 250

27Determinants in tourist expenditure composition

Regarding accommodation it seems that rented apartments and other typesof accommodation behave similarly and have the lowest budget share intransport In the case of hotels as the star rating increases expenditure on basiccompared to discretionary increases and expenditure on transport compared toat-destination decreases

As far as the length of stay is concerned longer stays increase the share ofnon-transport expenditure in basic expenses and reduce the share in transport

As regards destination and motivation tourists going to a seaside or country-side destination for leisure purposes behave quite differently from those goingto an urban destination for urban or cultural tourism The former group spendsmore on transport and less on the discretionary portion However those comingfor urban tourism show an increase in the share of non-transport expenses aswell as in the discretionary share compared to basic expenses

The total expenditure variable gives relevant information It is ordered alongthe first log-ratio axis from top to bottom meaning that those tourists havinga lower total expenditure spend relatively more on the transport share and thosewho have a higher total expenditure spend more on the other two parts of thebudget Regarding the expenditure made at destination per day it is ordereddiagonally from the upper-right corner to the bottom-left corner meaning thattourists with a higher at-destination expenditure per day spend less ontransportation and at the same time more on the discretionary part within at-destination expenses

Finally activities undertaken are located in the lower left quadrant of thegraph meaning that undertaking activities (discretionary expenses) decreasesthe share in transport expenses and although they continue to spend more onbasic expenses than on discretionary the discretionary share increases We havehighlighted some of them Those tourists attending sports and cultural eventsare those spending the most on discretionary compared to basic expensesHiking or some nautical sports could be free of charge thus tourists doing theseactivities have a basic to discretionary ratio close to the mean As for thetransportation share all activities tend to reduce this share about equally withthe exception of golf which leads to a much more substantial reduction

Conclusion and discussion

The main purpose of this article was to study the composition of touristexpenditure and its determinants that is the drivers of the share of touristexpenditure allocated to the different categories of a travel budget and withspecial emphasis on the distinction between legacy versus LCA travellers

The main differences between legacy and LCA users are that income somecountries of residence the travel group and some occupations have differenteffects depending on the type of airline Tourists with a medium or low incomewho travel with legacy airlines tend to spend a greater share of their travelbudget on transportation and basic expenses but those travelling with LCAstend to spend more at the destination and more specifically ondiscretionary expenditure LCA tourists residing in the lsquoother Europeancountriesrsquo spend more on transportation Families flying with legacy airlinesspend more on the transportation share compared to the at-destination share

TOURISM ECONOMICS28

and also spend more on basic expenses compared to discretionary ones Finallyin terms of occupational status legacy users with low-level employment spendmore on transportation whereas the opposite occurs with LCA travellers whospend more on the discretionary component Students and high-level employeesflying with legacy airlines spend more on the discretionary part of the tripbudget

As far as implications for management are concerned assuming that theobjectives of DMOs are to increase the allocation to the at-destinationcomponent of the trip budget (that is lowering the first log-ratio) some of therecommended actions would be to increase marketing efforts directed atmedium- and high-income earners especially at those travelling on LCAs toincrease marketing strategies in some of the European outbound markets suchas Germany (especially those using LCA) and Italy and the Netherlands(especially those flying with legacy airlines) DMOs should also focus on thosetourists travelling with friends (airline type is inconsequential) and studentsusing legacy companies As for the endogenous variables DMOs should directtheir efforts towards those tourists undertaking activities towards those stayinga little bit longer than a week and towards those with high total expenditureFurthermore if DMOs are also interested in capturing the groups who spendrelatively more on discretionary expenses (lowering the second log-ratio) theyshould focus their marketing efforts on those tourists who are highly educatedas well as those who have medium to high incomes and who use LCAsAdditionally they should focus on markets other than the UK and IrelandYoung tourists and tourists who travel with friends (where airline type isinconsequential) low-level employees and students using LCAs as well asstudents and high-level employees flying on legacy airlines should also betargeted In relation to the endogenous variables DMOs must take into accountthose tourists undertaking payable activities those on shorter stays or comingfor urban and cultural tourism or other purposes

The appeal of the CODA methodology for studying tourism budgets lies inthe fact that once the variables have been transformed the researcher can usestandard and well understood statistical models with unbounded error termdistributions while ensuring that the results will be compositionally coherentand that the standard statistical assumptions will hold Conversely results arenot compositionally coherent when we use the MANOVA model on raw sharewithout transforming it into log-ratios We have seen with our data that 84of individuals included in the sample have at least one limit of the 90 shareprediction intervals outside the [01] range

Contrary to the classic and highly restrictive alr and clr transformations theCODA methodology offers the potential to construct tailor-made log-ratios whichare intuitive to interpret and suit the research questions at hand A classificationtree of components is a clear and useful tool in this respect We encourageresearchers to use this approach in further research on tourism expenditure whenbudget share division is fundamental to the researchersrsquo questions

In this study we have encountered two limitations First contrary to studiesthat use macro data in micro data studies the coexistence of full-board halfboard bed and breakfast and accommodation-only tourists makes it impossibleto separate accommodation and food expenditure meaningfully Othersubdivisions (for example discretionary expenditure on activities non-grocery

29Determinants in tourist expenditure composition

shopping and moving around the destination) are technically feasible but canonly increase the percentages of zeros The CODA methodology involves somedegree of amalgamation of components Second even though there are someadvantages to using a database from an official statistics institution as in thiscase (large sample size and scope) the main disadvantage is that the set ofavailable variables cannot be controlled by the researcher A further issue to beaddressed is the time dimension As the EGATUR survey is conducted annuallyfurther research can be done to include a repeat cross-section analysis in orderto capture trends in the effects of predictors

References

Aitchison J (1986) The Statistical Analysis of Compositional Data Monographs on Statistics and AppliedProbability Chapman and Hall London (Reprinted in 2003 with additional material by TheBlackburn Press)

Alegre J Mateo S and Pou LL (2009) lsquoParticipation in tourism consumption and the intensityof participation an analysis of their socio-demographic and economic determinantsrsquo TourismEconomics Vol 15 No 3 pp 531ndash546

Bewley R (1982) lsquoThe generalized addilog demand system applied to Australian time series andcross section datarsquo Australian Economic Papers Vol 21 No 38 pp 177ndash192

Bewley R and Fiebig DG (1988) lsquoEstimation of price elasticities for an international telephonedemand modelrsquo The Journal of Industrial Economics Vol 36 No 4 pp 393ndash409

Cai LA (1998) lsquoAnalysing household food expenditure patterns on trips and vacations a Tobitmodersquo Journal of Hospitality and Tourism Research Vol 22 No 4 pp 338ndash358

Cai LA (1999) lsquoRelationship of household characteristics and lodging expenditure on leisure tripsrsquoJournal of Hospitality and Leisure Marketing Vol 6 No 2 pp 5ndash18

Cai LA Hong G-S and Morrison AM (1995) lsquoHousehold expenditure patterns for tourismproducts and servicesrsquo Journal of Travel and Tourism Marketing Vol 4 No 4 pp 15ndash39

Cannon TF and Ford J (2002) lsquoRelationship of demographic and trip characteristics to visitorspending an analysis of sports travel visitors across timersquo Tourism Economics Vol 8 No 3 pp 263ndash271

Castillo-Manzano JL and Marchena-Goacutemez M (2010) lsquoAnalysis of determinants of airline choiceprofiling the LCC passengerrsquo Applied Economics Letters Vol 18 No 1 pp 49ndash53

Chiou Y-C and Chen Y-H (2010) lsquoFactors influencing the intentions of passengers regarding full ser-vice and low cost carriers a notersquo Journal of Air Transport Management Vol 16 No 4 pp 226ndash228

Coenen M and van Eekeren L (2003) lsquoA study of the demand for domestic tourism by Swedishhouseholds using a two-staged budgeting modelrsquo Scandinavian Journal of Hospitality and TourismVol 3 No 2 pp 114ndash133

Craggs R and Schofield P (2009) lsquoExpenditure-based segmentation and visitor profiling at theQuays in Salford UKrsquo Tourism Economics Vol 15 No 1 pp 243ndash260

Deaton A and Muellbauer J (1980) lsquoAn almost ideal demand systemrsquo American Economic ReviewVol 70 No 3 pp 312ndash326

Dellaert B Borgers A and Timmermans H (1997) lsquoConsumer activity pattern choice devel-opment and test of stage-dependent conjoint choice experimentsrsquo Journal of Retailing and ConsumerServices Vol 4 No 1 pp 25ndash37

Divisekera S (2007) Modelling and Estimation of Tourism Demand Elasticities A Study of TourismExpenditure Allocation in Australia Cooperative Research Centre for Sustainable Tourism (CRCST)Gold Coast

Divisekera S (2009) lsquoEx-post demand for Australian tourism goods and servicesrsquo Tourism EconomicsVol 15 No 1 pp 153ndash180

Divisekera S (2010) lsquoEconomics of leisure and non-leisure tourist demand a study of domesticdemand for Australian tourismrsquo Tourism Economics Vol 16 No 1 pp 117ndash136

Dolnicar S Crouch GI Devinney T Huybers T Louviere JJ and Oppewal H (2008)lsquoTourism and discretionary income allocation Heterogeneity among householdsrsquo TourismManagement Vol 29 No 1 pp 44ndash52

Downward P and Lumsdon L (2000) lsquoThe demand for day-visits an analysis of visitor spendingrsquoTourism Economics Vol 6 No 3 pp 251ndash261

13Determinants in tourist expenditure composition

significant variables Age affects expenditure on meals as does travel groupwhich also affects transportation expenditure Marital status usually affects foodand accommodation expenditure

Since part expenditure in absolute terms is related to total expenditure acommon finding in these studies is that some of the explanatory variables affectall budget elements roughly equally For example Cai et al (1995) found thatthe higher the level of education the higher the expenditure in all budget partsand Wang et al (2006) conclude similarly about household income Such resultswill never be obtained when analysing budget share

The empirical analysis of budget share both for tourism expenditure andgeneral family budgets commonly involves estimating an almost ideal demandsystem of equations (Deaton and Muellbauer 1980) The almost ideal demandsystem is designed to analyse the interdependences of budget allocations thusovercoming the limitations of single-equation modelling and has receivedmuch attention in the last decade (Song et al 2012) It directly fits shares(composition) as dependent variables in a set of simultaneous regressions Theapproach has focused mainly on estimating price and income elasticities Studiesthat use macroeconomic data include among others those by Divisekera (20072009 2010) Fujii et al (1985) OrsquoHagan and Harrison (1984) Syriopoulos andSinclair (1993) and Wu et al (2011)

The almost ideal demand system has been applied both to data from a givenorigin to multiple destinations (that is ex ante ndash see Li et al 2004 Divisekera2009) or from multiple origins to a given destination which is the case in ourarticle (ex post ndash see Divisekera 2009) In ex post studies it is commonly assumedthat various commodities can be aggregated to broad bundles of productsprovided that prices in a bundle move in parallel and that the utility functionwith respect to tourism and other goods is weakly separable This makes itpossible to conceptualize tourism consumption as a multi-stage process In thefirst stage tourists allocate a household budget part to tourism consumptionin the second a tourism budget part to each tripdestination in the third adestination budget part to each good and service including transportationaccommodation and so on Ex post studies such as ours model only the laststage There are however studies of how tourism competes against othercategories of discretionary expenditure using individual micro data Forexample Melenberg and Van Soest (1996) use different parametric and semi-parametric Tobit models to explain the vacation budget share from householdcharacteristics and Dolnicar et al (2008) analyse how households allocatediscretionary income between tourism and competing uses

The review of Wang and Davidson (2010) concludes that since the vastmajority of tourism demand studies are conducted at macro level (this holdseven more for budget share analysis) there is room for more micro-econometricstudies in this area as the only manner of accounting for demand heterogeneityThe almost ideal demand system approach when applied to micro data caninclude individual characteristics For example Coenen and van Eekeren (2003)and Fleischer et al (2011) make ex ante studies of individual budget share inSweden and Israel respectively and use previous Heckit-type selectionequations to model the decision whether to travel or not Coenen and vanEekeren (2003) include household size and income as individual characteristicsFleischer et al (2011) add age education real state ownership place of birth

TOURISM ECONOMICS14

and Internet use The basic aim of those articles is to estimate elasticities andthey use the individual characteristics only as controls Fleischer et al (2011)are the only researchers to provide the estimates of equations predicting budgetshare from traveller characteristics as a by-product of their elasticity estimatesBeing born in the country of origin is reported to increase the transportationshare and reduce the share of on-site expenditures Education and householdreal estate assets reduce the share of on-site expenditures

The aim and approach of this article are similar to those of Coenen and vanEekeren (2003) and of Fleischer et al (2011) regarding the use of individualcharacteristics to predict budget share but differ in three important respects

bull Our study is ex post so that Heckit modelling is unfeasiblebull The effect of individual characteristics on budget composition (share) is the

core of the analysis Explanatory variables are individual characteristics ratherthan prices The results will include the effects of traveller heterogeneityrather than demand equations

bull Our analysis takes into account the compositional restrictions of the databy using the CODA methodology

The CODA methodology

Compared to absolute data compositional data such as budget share lie in aconstrained space A D-term composition measured on individual i xi1 xi2hellipxiD

has the following constraints

0 le xid le 1 and DΣ

d=1xid = 1 (1)

Aitchison (1986) and Pawlowsky-Glahn and Buccianti (2011) warn against theserious problems that arise when using standard statistical analysis tools oncompositional data Compositional data are non-normal and heteroscedasticOne component can increase only if some other(s) decreases This results innegative spurious correlations among the components and prevents interpretingeffects of linear models in the usual way lsquokeeping everything else constantrsquo

Even if specialized CODA techniques are starting to appear (for exampleRonning 1992 Thioacute-Henestrosa and Martiacuten-Fernaacutendez 2005) the easy way(Aitchison 1986 McLaren et al 1995 Fry et al 1996) involves transformingcompositional data so that they can be subject to standard and well-understoodstatistical techniques This is the approach we take in this article In short thisinvolves using the transformed share by means of logarithms of ratios insteadof the raw share

To ensure compositional coherence of predicted budget share (unit sum andnon-negativity) a set of parameter constraints is imposed to the almost idealdemand system However the presence of an error term with an unboundeddistribution (usually normal) results in a non-zero probability that actual sharelies outside the [01] interval (McLaren et al 1995 Fry et al 1996 Fry 2011)In other words the in fact bounded distribution of budget share results in amisspecification of the almost ideal demand system and of any model fittingpercentage share with an unbounded error distribution

15Determinants in tourist expenditure composition

The fact that the error term in any proper system of demand equationsapplied to budget share should take into account the data compositional naturehas been widely acknowledged (Aitchison 1986 Ronning 1992 Fry et al1996) However the review in Fry (2011) reports few studies that have appliedthe CODA methodology to demand equations and to the study of householdbudgets (McLaren et al 1995 Fry et al 1996 2000)

When fitting demand equations to log-ratios of expenditure componentsEngelrsquos aggregation condition which requires that household total expenditureshould equal the sum of components is automatically satisfied The model caneasily be extended to a full consumer demand analysis by the introduction ofprices and other covariates such as consumer characteristics (Aitchison 1986)McLaren et al (1995) relate the CODA analysis with log-ratio transformationto the almost ideal demand system and conclude that CODA makes it possibleto reach the same objectives with normal and homoscedastic error terms

Related developments are the indirect addilog system (Houthakker 1960)and the generalized addilog system (Bewley 1982) When applied tocompositions (for example Bewley and Fiebig 1988) the latter is equivalentto the CODA methodology in which the log-ratios of each component over thegeometric means of all components are the dependent variables in the set ofsimultaneous regressions

To the best of our knowledge there is no study of tourism budget share usingthe CODA methodology or any other methodology accounting for thecompositional constraints in compositional data

Method and data

Statistical approach

The simplest CODA approach involves applying standard statistical techniqueson logarithms of ratios of components Several log-ratio transformations havebeen suggested in the early CODA literature (Egozcue et al 2003) The additivelog-ratio transformation (alr) used by Fry et al (1996 2000) is the most popularand the easiest to compute given that it is simply the log-ratio of eachcomponent to the last

yid = ln(xidxiD) = ln(xid) ndash ln(xiD) with d = 123D ndash 1 (2)

The centred log-ratio transformation (clr) used by the generalized addilogdemand system (Bewley 1982) computes the log-ratios of each component overthe geometric mean of all the components including itself

⎛ xid ⎞yid = ln ⎜ ndashndashndashndashndashndashndashndashndashndashndashndash ⎟ with d = 123D (3)

⎝ D xi1xi2xi3 xiD ⎠

All log-ratio transformed yid variables recover the full unconstrained ndashinfin to infinrange It must be noted that one dimension is lost in the alr while in the clrone dimension is a linear combination of the remaining

TOURISM ECONOMICS16

The alr is commonly used for statistical modelling and prediction ofcompositions (Fry et al 1996) Conversely the clr transformation is commonlyused for statistical techniques which are based on a metric such as a clusteranalysis because of its preservation of distances even though it leads to asingular covariance matrix Thus while the alr would be appropriate for thepurpose of this article the fact that one component must be used as referencefor all others reduces its flexibility and interpretability Alternatives arepresented at the end of this subsection

While having zero expenditures in absolute data indeed has significantmethodological consequences (Lee 2001) these consequences are arguably moreserious in the CODA methodology If the xid variables contain zeros then log-ratios cannot be computed An obvious initial procedure to reduce zeros is toamalgamate small and conceptually similar components with many zeros intolarger ones In tourism budget research it can be useful to group together allexpenditure on activities or all expenditure on food for instance

In certain instances some zero components result from individual character-istics which are called essential zeros in the CODA literature (Aitchison 1986)Another typology of zeros encountered in the CODA literature is the roundingzero that is a component which is present but is too small to be detected bythe measurement instrument This is typical in chemical biological andgeological compositions and the CODA literature offers ample instruments todeal with rounding zeros

A classic essential zero example in economics is in household budget researchwhen measuring expenditure on tobacco and will essentially be zero if allmembers are non-smokers If tobacco expenditure is decidedly in the researchersrsquointerest the target population should be redefined to include only smokers Inmany instances budget research is often not clear whether zero expenditurescome closer to being essential or rounding zeros In some cases they can beunderstood as corner solutions in a utility maximization problem In others theycan be understood as the inherent randomness of human behaviour or as thelimitations of the data Tourists may spend a certain amount on activities oncertain trips but not on others and so surveys of only one trip will unavoidablycontain some zeros of this type Tourists may also forget or fail to report trivialexpenses such as postcard shopping local bus tickets and going to a museum(see Legoheacuterel (1998) for a discussion on the instability of tourism expenditure)Fry et al (2000) claim that in both situations zeros can be proxied by a verysmall value and thus be treated as rounding zeros In tourist budget researchto treat zero expenses in activities as rounding zeros implies assuming firstthat there is basically no tourist type who will never spend anything onactivities and second that tourists who generally spend little on activities arebasically similar to those who spend nothing or fail to report small expensesWe find both assumptions to be reasonable

Fry et al (2000) essentially used the same zero replacement strategy that waslater suggested by Martiacuten-Fernaacutendez et al (2003) namely replacing xid = 0 with

xprimeid = kδid with 0 lt k lt 1 (4)

where δid is the smallest detectable proportion for individual i and component d

17Determinants in tourist expenditure composition

Martiacuten-Fernaacutendez et al (2003) suggest using k = 065 although a sensitivityanalysis of the results on the choice of k is always advisable (we used k = 030and k = 099 with no sizeable change in the estimates) Next non-zero xid valueshave to be reduced in order to preserve the unit sum and the ratios among non-zero components As suggested by Martiacuten-Fernaacutendez et al (2003) with

xprimeid = xid (1 ndash Σxid=0

xprimeid) (5)

Simulations show this method performs particularly well if the proportion ofzeros is below 10 (Martiacuten-Fernaacutendez et al 2011)

In our data set zeros were present only in one budget category (discretionaryexpenditure) The minimum amount spent by the non-zero group was euro1which roughly corresponds to the price of a city bus ticket the entrance to asubsidized local museum or a cheap souvenir Since the total expenditureis known for each individual we compute δid by dividing euro1 with the totalexpenditure of individual I (see the appendix for the SPSS commandsyntax)

In this article we consider alternative log-ratio transformations which aremore flexible than the alr and clr in that the denominator does not have tobe the same in all ratios This increased flexibility makes it easier to computelog-ratios which are more interpretable with respect to the researchersrsquo questions orhypotheses

In general an interpretable log-ratio transformation is easy to computewhenever there is an interpretable sequential binary partition of componentsinto pairs of groups of components according to the researchersrsquo objectives orto the conceptual similarity of the components These partitions start bydividing components into two clusters and then continue by subdividingone of the clusters into two until each component constitutes its owncluster D components always involve Dndash1 partitions These partitions are bestunderstood as a partition tree or dendrogram (Pawlowsky-Glahn and Egozcue2011)

A meaningful log-ratio transformation takes ratios of the geometric meansof the two component clusters at each partition Numerators and denominatorsare interchangeable In our article we consider xi1 = transportation expenditurexi2 = accommodation and food (basic expenditure) and xi3 = activities andshopping (discretionary expenditure) An interpretable sequential partition isshown in Figure 1

The sequential partition in Figure 1 can have a dual interpretation It mayimply that researchers consider both types of at-destination expenses to bemutually similar and less similar to transportation Or it may imply that theresearch questions involve the distribution of total expenditure betweentransportation and at-destination expenditure and the distribution ofat-destination expenditure into basic and discretionary components as is thecase in our article

The first log-ratio compares transportation expenditure with the geometricmean of accommodation and food (basic expenditure) and activities andshopping (discretionary expenditure) With this ratio we want to observe howtravelling with an LCA or a legacy company affects the share of transportationcompared to non-transportation (at-destination) expenses

TOURISM ECONOMICS18

Figure 1 Sequential partition of total expenditure

⎛ xi1 ⎞ 1 1yi1 = ln ⎜ ndashndashndashndash ⎟ = ln(xi1) ndash ndash ln(xi2) ndash ndash ln(xi3) (6)

⎝ xi2xi3 ⎠ 2 2

Positive values show a transportation share greater than the geometric meanof the remaining two components Negative values show the opposite

The second log-ratio is a ratio of accommodation and food (basic expenditure)over activities and shopping (discretionary expenditure) With this ratio wewant to find out to what tourists allocate the rest of their trip budget andwhether more is allocated to basic or discretionary expenditure once they havepaid for (often in advance) their transportation

⎛ xi2 ⎞yi2 = ln ⎜ ndashndashndash ⎟ = ln(xi2) ndash ln(xi3) (7)

⎝ xi3 ⎠

Positive values show a basic (accommodation and food) share which is largerthan the discretionary (activities and shopping) share Negative values illustratethe opposite

Log-ratio transformations based on sequential binary partitions areproportional to the isometric log-ratio transformation (ilr see Egozcue et al2003 Egozcue and Pawlowsky-Glahn 2005)

As compositions are vector variables they cannot be analysed component-wise by means of univariate regression ANOVA models and the likeSeemingly unrelated regression models for continuous explanatory variables ormultivariate ANOVA (MANOVA) models for categorical explanatory variablesare appropriate MANOVArsquos multivariate tests and statistics (for example

x2 x3 Transportation

Global partition of total expenditure

x1

Accommodation

amp food

(basic)

Activities

amp shopping

(discretionary)

Partition of

expenses other

than transportation

(at-destination

expenses)

19Determinants in tourist expenditure composition

Pillairsquos trace Hotellingrsquos trace or Wilkrsquos lambda) are invariant to howcomponents are arranged in Figure 1 to compute log-ratios (Mateu-Figueras etal 2013) Univariate tests referring to each particular log-ratio are not invari-ant hence the importance of the interpretability of each log-ratio The softwareSPSS 19 is used to estimate the MANOVA model (GLM procedure)

We have to distinguish between three types of variables moderatingexogenous and endogenous to be included in the analysis

Moderating variables are those which modify the effect of exogenousvariables In our article we treat type of airline as a moderator in other wordswe include its interaction terms with all other variables in the MANOVA model

Exogenous variables are assumed to affect expenditure but not the converseSocio-demographic characteristics (travellerrsquos age gender education incomeetc) are obviously determined prior to any travel decision and hence exogenouswith respect to expenditure

Many of the choices tourists make are interdependent (Dellaert et al 1997)or at least planned simultaneously (Fesenmaier and Jeng 2000) So expenditureis arguably decided on at the same time as many trip attributes At least itis unclear whether expenditure is consistently decided on after trip attributesby all travellers We consider them to be endogenous variables and thereforedo not include them in the MANOVA model as explanatory in order to preventendogeneity problems Instead we display them graphically in the log-ratiospace The means of both log-ratios within each endogenous category are thecategory coordinates

The large sample size (see next subsection) makes it possible to use lowp-values Moderating effects with p-values higher than 001 according to eitherPillairsquos trace Hotellingrsquos trace or Wilksrsquo lambda were removed from the modelAll the variables and moderating effects included in the final model aresignificant at 001

Sample and variables

In this article we use secondary official statistics data The data were providedby the Instituto de Estudios Turiacutesticos (IET) an official agency of the Ministryof Industry Energy and Tourism and one which produces the majority oftourism data in Spain The survey is known as the Encuesta de Gasto Turiacutestico(EGATUR) in which tourism expenditure and other tourist information suchas trip information and tourist sociodemographic characteristics are studiedThe EGATUR survey conducted in 27 major Spanish airports in 2010 usedCAPI (computer assisted personal interview) to interview tourists leaving thecountry The sample is non-proportionally stratified by country of residenceairport and month See IET (2012) for further details on the EGATURmethodology

Our universe is a subset of the EGATUR universe which consists ofEuropean leisure visitors arriving by air and spending at least one night inSpain We excluded flights from outside Europe because LCAs mostly operateshort-haul flights We also centred our study on only those trips with one singledestination thus excluding multi-stage trips as the decision process regardingexpenditure composition for these trips is expected to fundamentally differ fromthat of single-stage trips Stays of over 120 days were also excluded

TOURISM ECONOMICS20

For this study we did not consider tourists

bull who have essential zeros in accommodation (tourists who own a house at thedestination or tourists who stay with friends or relatives)

bull who do not decide how much they will spend on certain components(business and study trips)

bull who do not pay for the trip themselves (trips paid for by employers familyfriends prizes offers etc)

bull for whom composition is wholly or partly unobserved (package tourists)

The final sample size was N= 19359From the expenditure variables included in the EGATUR survey database

and used in the model as budget components we first put the amount paidfor transportation (x1) This component has no zeros

Second the amount of money paid for accommodation and food isundistinguishable for full-board half-board or bed and breakfast accommodationTherefore we define a joint accommodation and food component (basic expendi-ture) In this component we include not only the amount paid for consumptionin bars and restaurants but also for buying groceries and everyday products insupermarkets (x2) This component has no zeros

Finally EGATUR provides an aggregated expenditure on activities andshopping (except groceries and everyday products) To this we added theconceptually similar amount paid for moving around the destination (publictransportation andor car rented at the destination) in order to build anactivities and shopping component (x3) This component had 98 zeros whichstresses its discretionary character Zeros were replaced as explained in thestatistical approach subsection

We consider as at-destination (or also called non-transportation) expenses thebasic component (food and accommodation) and the discretionary component(activities and shopping)

Share and log-ratios are described in Table 1 Explanatory variables are thelevel of education income country of residence gender travel groupprofessional status (for pensioners we consider their last professional position)and age The age variable is built as a combination of the original age variableand the variable referring to the individual economic situation This is donein order to have a specific pensioner category taking into account the varyingretirement ages across countries and professions and thus capturing thosetourists who have more free time to undertake a trip regardless of physical ageTable 2 shows their categories and frequencies

As endogenous variables which are not included in the MANOVA modelbut are included in the log-ratio space we include activities undertaken at thedestination accommodation length of stay total expenditure quartilesquartiles of expenditure made at destination per day (basic plus discretionaryexpenses) and a combination of motivation and destination This last variablehas been constructed as a combination of the motivation and the destinationvariable We distinguish between first those tourists travelling for culturaltourism to singular cities second those who come just for leisure either inthe countryside or more commonly at the seaside and third other typologies(Table 3)

21Determinants in tourist expenditure composition

Table 1 Percentage share and log-ratio descriptive statistics

Min Max Mean SD Skewness Kurtosis

Transportation component (x1) 003 9611 2846 1387 0837 0789Basic component (x2) 021 9963 5472 1613 ndash0114 ndash0450Discretionary component (x3) 001 8668 1681 1292 0988 1174Transportationat-destinationlog-ratio (y1) ndash757 548 01506 10702 0714 1404

Basicdiscretionary log-ratio (y2) ndash530 872 17657 18173 1573 2151

Results

Results for the exogenous variables

Table 4 presents the MANOVA results The univariate R2 corrected for thedegrees of freedom is 0045 (first log-ratio) and 0105 (second log-ratio) Themultivariate uncorrected R2 is 1 ndash Λ = 0178 and corrected for the degrees offreedom is 0176 We checked that the results did not change when removingthe four outliers with the highest Cookrsquos distance

Table 4 presents the estimates for the two log-ratios the log-ratio oftransportation over the other two categories (y1) and the log-ratio of basicexpenditure (accommodation and food) over discretionary expenditure (activitiesand shopping) (y2) respectively Because of the moderating effects within eachlog-ratio there are two columns representing the effects when flying with anLCA or with a legacy airline (main and moderating effects have already addedtogether for easier reading) If the LCA and legacy columns are different thereis a significant moderating effect (p-value lt 001)

As we are interpreting log-ratios a positive estimate of a given predictorcategory means that tourists in that category spend more on the numeratorcompared to the denominator than tourists in the reference predictor categoryA negative estimate shows the opposite

The intercept term shows the main effect of company type that is thepredicted log-ratio for each type of airline within the reference category for allvariables (university education medium income resident in the UK or Ireland25ndash44 yearsrsquo old male travelling with partner and mid-level employee) Withinthe reference categories legacy users spend a higher proportion on transportationcompared to other expenses and LCA users spend more on basic expensescompared to discretionary

The level of education seems to have more to do with activities undertakenthan with transportation Results show that a lower level of education resultsin higher expenditure on basic expenses compared to discretionary and is almostequal for users of both types of airline For mainly or only LCA users a lowerlevel of education results in a higher share in transport compared to the othertwo categories If we put it in another way there seems to be a distinct highlyeducated LCA user segment which utilizes the savings in transport toincrease expenditure in non-transportation expenses and even more so indiscretionary

TOURISM ECONOMICS22

Table 2 Frequency distributions of exogenous and moderating variables in the MANOVAmodel

Variables Count Percentage

Level of educationUp to high school 6019 311Universitya 13340 689

Income categoryMediumlow 797 41Mediuma 13186 681Mediumhigh 4376 226High 1000 52

Country of residencePortugal 624 32Other European countries 850 44Belgium 1006 52Netherlands 1069 55Austria Switzerland and Liechtenstein 1122 58France 1326 68Scandinavian countries 1789 92Germany 2363 122Italy 3113 181United Kingdom and Irelanda 6097 315

GenderFemale 8530 441Malea 10829 559

Travel groupAlone 2484 128In family 2980 154With friends 4440 229With partnera 9446 488

Professional statusHomemaker 517 27Unemployed 602 31Student 1432 74Low-level employee 915 47Mid-level employeea 10964 586High-level employee 2138 110Self-employed 2791 144

AgeOver 45 yearsrsquo old and pensioner 1484 77Over 45 yearsrsquo old and not pensioner 5009 25925ndash44 yearsrsquo olda 10569 54615ndash24 yearsrsquo old 2297 119

Type of airline (moderating variable)Legacy 5884 304Low cost 13475 696

Note aReference categories in the MANOVA model chosen either because they are the largestcategories or because they are the most standard conceptually considered

23Determinants in tourist expenditure composition

Table 3 Frequency distributions of endogenous variables (not included in the MANOVAmodel)

Variables Count Percentage

ActivitiesGolfYes 227 12No 19132 988HikingYes 364 19No 18995 981Sporting eventsYes 603 31No 18756 969Nautical sportsYes 1645 85No 17714 915Other sportsYes 904 47No 18455 953Cultural visitsYes 11180 578No 8179 422Cultural eventsYes 2452 127No 16907 873Other cultural activitiesYes 4175 216No 15184 784SpaYes 1199 62No 18160 938Theme parksYes 1745 90No 17614 910GastronomyYes 1565 81No 17794 919Type of accommodationHotel 4ndash5 5056 261Hotel 3 4922 254Hotel lt 3 6673 345Rented appartment 2116 109Other accommodation 592 31Length of stay5 or less nights 9911 5126ndash8 nights 5952 3079ndash12 nights 1592 8213ndash15 nights 1261 6516 and more nights 643 33

TOURISM ECONOMICS24

Table 3 continued

Variables Count Percentage

Total expenditureQuartile 1 5074 262Quartile 2 5104 264Quartile 3 4986 258Quartile 4 4195 217Expenditure at destination per dayQuartile 1 4855 251Quartile 2 4832 250Quartile 3 4868 251Quartile 4 4804 248Motivation-destinationLeisure on coastin countryside 8083 418Urban tourism 5324 275Others 5952 307

Level of income in contrast affects both log-ratios The higher the incomethe lower the transport share The higher the income the lower the basicdiscretionary expense ratio that is a higher portion of the non-transportexpense is devoted to the discretionary budget

As far as the country of residence is concerned the results show that it affectsmainly the basicdiscretionary log-ratio and exhibits strong airline typemoderating effects LCA users tend to differ depending on their country ofresidence when distributing non-transport expenses into basic versusdiscretionary In general when compared to the UK and Ireland residentcategory (reference category) all other European residents have a largerdiscretionary share More precisely the French Italians and Portuguese have thehighest discretionary share within LCA users and other European countries(mainly central and eastern European) have the highest discretionary sharewithin legacy airline users

Gender has a small effect which is constant for both types of airline Femaletravellers tend to slightly increase the transport share and the share of basicexpenses within the non-transport expenses

The travel group seems to affect both log-ratios and has a moderating effectwith airline type Those tourists who travel in a family group have a highershare in transport expenses when flying with legacy airlines In the second log-ratio travelling as a family with legacy airlines increases the basic expensesshare compared to the share in discretionary Travelling with friends increasesthe share in non-transport expenses for both airline types Besides those whotravel with friends spend more in the discretionary share compared to basicexpenses and those who fly with LCA even more so Finally travelling aloneincreases the basic expenses share compared to the share in discretionaryexpenses for both airline types

When professional status is considered the results show that it mostly affectsthe basicdiscretionary log-ratio and there are some relevant airline type

25Determinants in tourist expenditure composition

Table 4 MANOVA results

y1 log-ratio y2 log-ratiotransportation basicat-destination discretionary

LCA Legacy LCA Legacy

Intercept 0225 0274 2195 1881

Up to high school 0192 0078 0406 0437University 0 0 0 0

Mediumlow income 0073 0289 0037 0518Medium income 0 0 0 0Mediumhigh income ndash0227 ndash0058 ndash0269 ndash0006High income ndash0495 ndash0190 ndash0463 ndash0141

Residence in Portugal 0041 0069 ndash1136 ndash0580Residence in other European Countries 0326 ndash0012 ndash0772 ndash0820Residence in Belgium ndash0127 ndash0159 ndash0667 ndash0461Residence in Netherlands ndash0120 ndash0197 ndash0678 ndash0530Residence in Austria + Switzerland +Liechtenstein ndash0017 ndash0087 ndash0687 ndash0436

Residence in France ndash0112 ndash0111 ndash0870 ndash0553Residence in Scandinavian countries ndash0045 ndash0038 ndash0660 ndash0477Residence in Germany ndash0283 ndash0176 ndash0452 ndash0392Residence in Italy ndash0209 ndash0266 ndash0855 ndash0503Residence in UK or Ireland 0 0 0 0

Female 0094 0094 0093 0093Male 0 0 0 0

Travelling alone ndash0053 ndash0002 0081 0126Travelling in family 0085 0193 0014 0213Travelling with friends ndash0237 ndash0191 ndash0504 ndash0282Travelling with partner 0 0 0 0

Homemaker 0080 0160 0253 0235Unemployed 0030 0159 ndash0036 ndash0039Student ndash0005 ndash0104 ndash0331 ndash0624Low-level employee ndash0034 0194 ndash0194 0063Mid-level employee 0 0 0 0High-level employee ndash0008 ndash0064 ndash0001 ndash0139Self-employed 0029 ndash0086 0193 ndash0072

Over 45 yearsrsquo old and pensioner ndash0060 ndash0060 0633 0633Over 45 yearsrsquo old and not pensioner ndash0010 ndash0010 0142 014225ndash44 yearsrsquo old 0 0 0 015ndash24 yearsrsquo old 0002 0002 ndash0019 ndash0019

moderating effects Legacy users with a low professional status (homemakerunemployed or low-level employee) spend more on transport compared to thenon-transport expenses Conversely students using legacy airlines spend moreon non-transport expenses and these non-transport expenses are to a greater

TOURISM ECONOMICS26

Figure 2 Endogenous variables on the log-ratio space Mean log-ratio valueswithin each category of the endogenous variables

extent devoted to discretionary expenses compared to basic expensesHomemakers increase basic expenditure compared to discretionary with respectto the reference category (mid-level employee) for both airline types LCA userswho are self-employed also increase basic expenditure compared to discretionaryLow-level employees flying with an LCA increase the discretionary sharecompared to the basic and high-level employees flying with legacy airlines alsospend relatively more on the discretionary part

The results show that age affects the second log-ratio more than the firstand there is no moderating effect with the airline type Being a pensioner resultsin a higher share in basic expenses compared to discretionary expenses for bothairline types The remaining categories have hardly any differences from thereference 25ndash44 yearsrsquo old

Results for the endogenous variables

As introduced in the methodology section Figure 2 shows how the endogenousvariables behave in the log-ratio space The origin of the graph is representedby the mean log-ratios in Table 1 01506 for the transportother ratio and17657 for the basicdiscretionary ratio and according to this the graph showscategories with higher or lower ratios than the average

075

060

045

030

015

000

ndash015

ndash030

ndash045

ndash060

Ratio tra

nsport

ationa

t-destination e

xpenses

Ratio basicdiscretionary expenses

050 075 100 125 150 175 200 225 250

27Determinants in tourist expenditure composition

Regarding accommodation it seems that rented apartments and other typesof accommodation behave similarly and have the lowest budget share intransport In the case of hotels as the star rating increases expenditure on basiccompared to discretionary increases and expenditure on transport compared toat-destination decreases

As far as the length of stay is concerned longer stays increase the share ofnon-transport expenditure in basic expenses and reduce the share in transport

As regards destination and motivation tourists going to a seaside or country-side destination for leisure purposes behave quite differently from those goingto an urban destination for urban or cultural tourism The former group spendsmore on transport and less on the discretionary portion However those comingfor urban tourism show an increase in the share of non-transport expenses aswell as in the discretionary share compared to basic expenses

The total expenditure variable gives relevant information It is ordered alongthe first log-ratio axis from top to bottom meaning that those tourists havinga lower total expenditure spend relatively more on the transport share and thosewho have a higher total expenditure spend more on the other two parts of thebudget Regarding the expenditure made at destination per day it is ordereddiagonally from the upper-right corner to the bottom-left corner meaning thattourists with a higher at-destination expenditure per day spend less ontransportation and at the same time more on the discretionary part within at-destination expenses

Finally activities undertaken are located in the lower left quadrant of thegraph meaning that undertaking activities (discretionary expenses) decreasesthe share in transport expenses and although they continue to spend more onbasic expenses than on discretionary the discretionary share increases We havehighlighted some of them Those tourists attending sports and cultural eventsare those spending the most on discretionary compared to basic expensesHiking or some nautical sports could be free of charge thus tourists doing theseactivities have a basic to discretionary ratio close to the mean As for thetransportation share all activities tend to reduce this share about equally withthe exception of golf which leads to a much more substantial reduction

Conclusion and discussion

The main purpose of this article was to study the composition of touristexpenditure and its determinants that is the drivers of the share of touristexpenditure allocated to the different categories of a travel budget and withspecial emphasis on the distinction between legacy versus LCA travellers

The main differences between legacy and LCA users are that income somecountries of residence the travel group and some occupations have differenteffects depending on the type of airline Tourists with a medium or low incomewho travel with legacy airlines tend to spend a greater share of their travelbudget on transportation and basic expenses but those travelling with LCAstend to spend more at the destination and more specifically ondiscretionary expenditure LCA tourists residing in the lsquoother Europeancountriesrsquo spend more on transportation Families flying with legacy airlinesspend more on the transportation share compared to the at-destination share

TOURISM ECONOMICS28

and also spend more on basic expenses compared to discretionary ones Finallyin terms of occupational status legacy users with low-level employment spendmore on transportation whereas the opposite occurs with LCA travellers whospend more on the discretionary component Students and high-level employeesflying with legacy airlines spend more on the discretionary part of the tripbudget

As far as implications for management are concerned assuming that theobjectives of DMOs are to increase the allocation to the at-destinationcomponent of the trip budget (that is lowering the first log-ratio) some of therecommended actions would be to increase marketing efforts directed atmedium- and high-income earners especially at those travelling on LCAs toincrease marketing strategies in some of the European outbound markets suchas Germany (especially those using LCA) and Italy and the Netherlands(especially those flying with legacy airlines) DMOs should also focus on thosetourists travelling with friends (airline type is inconsequential) and studentsusing legacy companies As for the endogenous variables DMOs should directtheir efforts towards those tourists undertaking activities towards those stayinga little bit longer than a week and towards those with high total expenditureFurthermore if DMOs are also interested in capturing the groups who spendrelatively more on discretionary expenses (lowering the second log-ratio) theyshould focus their marketing efforts on those tourists who are highly educatedas well as those who have medium to high incomes and who use LCAsAdditionally they should focus on markets other than the UK and IrelandYoung tourists and tourists who travel with friends (where airline type isinconsequential) low-level employees and students using LCAs as well asstudents and high-level employees flying on legacy airlines should also betargeted In relation to the endogenous variables DMOs must take into accountthose tourists undertaking payable activities those on shorter stays or comingfor urban and cultural tourism or other purposes

The appeal of the CODA methodology for studying tourism budgets lies inthe fact that once the variables have been transformed the researcher can usestandard and well understood statistical models with unbounded error termdistributions while ensuring that the results will be compositionally coherentand that the standard statistical assumptions will hold Conversely results arenot compositionally coherent when we use the MANOVA model on raw sharewithout transforming it into log-ratios We have seen with our data that 84of individuals included in the sample have at least one limit of the 90 shareprediction intervals outside the [01] range

Contrary to the classic and highly restrictive alr and clr transformations theCODA methodology offers the potential to construct tailor-made log-ratios whichare intuitive to interpret and suit the research questions at hand A classificationtree of components is a clear and useful tool in this respect We encourageresearchers to use this approach in further research on tourism expenditure whenbudget share division is fundamental to the researchersrsquo questions

In this study we have encountered two limitations First contrary to studiesthat use macro data in micro data studies the coexistence of full-board halfboard bed and breakfast and accommodation-only tourists makes it impossibleto separate accommodation and food expenditure meaningfully Othersubdivisions (for example discretionary expenditure on activities non-grocery

29Determinants in tourist expenditure composition

shopping and moving around the destination) are technically feasible but canonly increase the percentages of zeros The CODA methodology involves somedegree of amalgamation of components Second even though there are someadvantages to using a database from an official statistics institution as in thiscase (large sample size and scope) the main disadvantage is that the set ofavailable variables cannot be controlled by the researcher A further issue to beaddressed is the time dimension As the EGATUR survey is conducted annuallyfurther research can be done to include a repeat cross-section analysis in orderto capture trends in the effects of predictors

References

Aitchison J (1986) The Statistical Analysis of Compositional Data Monographs on Statistics and AppliedProbability Chapman and Hall London (Reprinted in 2003 with additional material by TheBlackburn Press)

Alegre J Mateo S and Pou LL (2009) lsquoParticipation in tourism consumption and the intensityof participation an analysis of their socio-demographic and economic determinantsrsquo TourismEconomics Vol 15 No 3 pp 531ndash546

Bewley R (1982) lsquoThe generalized addilog demand system applied to Australian time series andcross section datarsquo Australian Economic Papers Vol 21 No 38 pp 177ndash192

Bewley R and Fiebig DG (1988) lsquoEstimation of price elasticities for an international telephonedemand modelrsquo The Journal of Industrial Economics Vol 36 No 4 pp 393ndash409

Cai LA (1998) lsquoAnalysing household food expenditure patterns on trips and vacations a Tobitmodersquo Journal of Hospitality and Tourism Research Vol 22 No 4 pp 338ndash358

Cai LA (1999) lsquoRelationship of household characteristics and lodging expenditure on leisure tripsrsquoJournal of Hospitality and Leisure Marketing Vol 6 No 2 pp 5ndash18

Cai LA Hong G-S and Morrison AM (1995) lsquoHousehold expenditure patterns for tourismproducts and servicesrsquo Journal of Travel and Tourism Marketing Vol 4 No 4 pp 15ndash39

Cannon TF and Ford J (2002) lsquoRelationship of demographic and trip characteristics to visitorspending an analysis of sports travel visitors across timersquo Tourism Economics Vol 8 No 3 pp 263ndash271

Castillo-Manzano JL and Marchena-Goacutemez M (2010) lsquoAnalysis of determinants of airline choiceprofiling the LCC passengerrsquo Applied Economics Letters Vol 18 No 1 pp 49ndash53

Chiou Y-C and Chen Y-H (2010) lsquoFactors influencing the intentions of passengers regarding full ser-vice and low cost carriers a notersquo Journal of Air Transport Management Vol 16 No 4 pp 226ndash228

Coenen M and van Eekeren L (2003) lsquoA study of the demand for domestic tourism by Swedishhouseholds using a two-staged budgeting modelrsquo Scandinavian Journal of Hospitality and TourismVol 3 No 2 pp 114ndash133

Craggs R and Schofield P (2009) lsquoExpenditure-based segmentation and visitor profiling at theQuays in Salford UKrsquo Tourism Economics Vol 15 No 1 pp 243ndash260

Deaton A and Muellbauer J (1980) lsquoAn almost ideal demand systemrsquo American Economic ReviewVol 70 No 3 pp 312ndash326

Dellaert B Borgers A and Timmermans H (1997) lsquoConsumer activity pattern choice devel-opment and test of stage-dependent conjoint choice experimentsrsquo Journal of Retailing and ConsumerServices Vol 4 No 1 pp 25ndash37

Divisekera S (2007) Modelling and Estimation of Tourism Demand Elasticities A Study of TourismExpenditure Allocation in Australia Cooperative Research Centre for Sustainable Tourism (CRCST)Gold Coast

Divisekera S (2009) lsquoEx-post demand for Australian tourism goods and servicesrsquo Tourism EconomicsVol 15 No 1 pp 153ndash180

Divisekera S (2010) lsquoEconomics of leisure and non-leisure tourist demand a study of domesticdemand for Australian tourismrsquo Tourism Economics Vol 16 No 1 pp 117ndash136

Dolnicar S Crouch GI Devinney T Huybers T Louviere JJ and Oppewal H (2008)lsquoTourism and discretionary income allocation Heterogeneity among householdsrsquo TourismManagement Vol 29 No 1 pp 44ndash52

Downward P and Lumsdon L (2000) lsquoThe demand for day-visits an analysis of visitor spendingrsquoTourism Economics Vol 6 No 3 pp 251ndash261

TOURISM ECONOMICS14

and Internet use The basic aim of those articles is to estimate elasticities andthey use the individual characteristics only as controls Fleischer et al (2011)are the only researchers to provide the estimates of equations predicting budgetshare from traveller characteristics as a by-product of their elasticity estimatesBeing born in the country of origin is reported to increase the transportationshare and reduce the share of on-site expenditures Education and householdreal estate assets reduce the share of on-site expenditures

The aim and approach of this article are similar to those of Coenen and vanEekeren (2003) and of Fleischer et al (2011) regarding the use of individualcharacteristics to predict budget share but differ in three important respects

bull Our study is ex post so that Heckit modelling is unfeasiblebull The effect of individual characteristics on budget composition (share) is the

core of the analysis Explanatory variables are individual characteristics ratherthan prices The results will include the effects of traveller heterogeneityrather than demand equations

bull Our analysis takes into account the compositional restrictions of the databy using the CODA methodology

The CODA methodology

Compared to absolute data compositional data such as budget share lie in aconstrained space A D-term composition measured on individual i xi1 xi2hellipxiD

has the following constraints

0 le xid le 1 and DΣ

d=1xid = 1 (1)

Aitchison (1986) and Pawlowsky-Glahn and Buccianti (2011) warn against theserious problems that arise when using standard statistical analysis tools oncompositional data Compositional data are non-normal and heteroscedasticOne component can increase only if some other(s) decreases This results innegative spurious correlations among the components and prevents interpretingeffects of linear models in the usual way lsquokeeping everything else constantrsquo

Even if specialized CODA techniques are starting to appear (for exampleRonning 1992 Thioacute-Henestrosa and Martiacuten-Fernaacutendez 2005) the easy way(Aitchison 1986 McLaren et al 1995 Fry et al 1996) involves transformingcompositional data so that they can be subject to standard and well-understoodstatistical techniques This is the approach we take in this article In short thisinvolves using the transformed share by means of logarithms of ratios insteadof the raw share

To ensure compositional coherence of predicted budget share (unit sum andnon-negativity) a set of parameter constraints is imposed to the almost idealdemand system However the presence of an error term with an unboundeddistribution (usually normal) results in a non-zero probability that actual sharelies outside the [01] interval (McLaren et al 1995 Fry et al 1996 Fry 2011)In other words the in fact bounded distribution of budget share results in amisspecification of the almost ideal demand system and of any model fittingpercentage share with an unbounded error distribution

15Determinants in tourist expenditure composition

The fact that the error term in any proper system of demand equationsapplied to budget share should take into account the data compositional naturehas been widely acknowledged (Aitchison 1986 Ronning 1992 Fry et al1996) However the review in Fry (2011) reports few studies that have appliedthe CODA methodology to demand equations and to the study of householdbudgets (McLaren et al 1995 Fry et al 1996 2000)

When fitting demand equations to log-ratios of expenditure componentsEngelrsquos aggregation condition which requires that household total expenditureshould equal the sum of components is automatically satisfied The model caneasily be extended to a full consumer demand analysis by the introduction ofprices and other covariates such as consumer characteristics (Aitchison 1986)McLaren et al (1995) relate the CODA analysis with log-ratio transformationto the almost ideal demand system and conclude that CODA makes it possibleto reach the same objectives with normal and homoscedastic error terms

Related developments are the indirect addilog system (Houthakker 1960)and the generalized addilog system (Bewley 1982) When applied tocompositions (for example Bewley and Fiebig 1988) the latter is equivalentto the CODA methodology in which the log-ratios of each component over thegeometric means of all components are the dependent variables in the set ofsimultaneous regressions

To the best of our knowledge there is no study of tourism budget share usingthe CODA methodology or any other methodology accounting for thecompositional constraints in compositional data

Method and data

Statistical approach

The simplest CODA approach involves applying standard statistical techniqueson logarithms of ratios of components Several log-ratio transformations havebeen suggested in the early CODA literature (Egozcue et al 2003) The additivelog-ratio transformation (alr) used by Fry et al (1996 2000) is the most popularand the easiest to compute given that it is simply the log-ratio of eachcomponent to the last

yid = ln(xidxiD) = ln(xid) ndash ln(xiD) with d = 123D ndash 1 (2)

The centred log-ratio transformation (clr) used by the generalized addilogdemand system (Bewley 1982) computes the log-ratios of each component overthe geometric mean of all the components including itself

⎛ xid ⎞yid = ln ⎜ ndashndashndashndashndashndashndashndashndashndashndashndash ⎟ with d = 123D (3)

⎝ D xi1xi2xi3 xiD ⎠

All log-ratio transformed yid variables recover the full unconstrained ndashinfin to infinrange It must be noted that one dimension is lost in the alr while in the clrone dimension is a linear combination of the remaining

TOURISM ECONOMICS16

The alr is commonly used for statistical modelling and prediction ofcompositions (Fry et al 1996) Conversely the clr transformation is commonlyused for statistical techniques which are based on a metric such as a clusteranalysis because of its preservation of distances even though it leads to asingular covariance matrix Thus while the alr would be appropriate for thepurpose of this article the fact that one component must be used as referencefor all others reduces its flexibility and interpretability Alternatives arepresented at the end of this subsection

While having zero expenditures in absolute data indeed has significantmethodological consequences (Lee 2001) these consequences are arguably moreserious in the CODA methodology If the xid variables contain zeros then log-ratios cannot be computed An obvious initial procedure to reduce zeros is toamalgamate small and conceptually similar components with many zeros intolarger ones In tourism budget research it can be useful to group together allexpenditure on activities or all expenditure on food for instance

In certain instances some zero components result from individual character-istics which are called essential zeros in the CODA literature (Aitchison 1986)Another typology of zeros encountered in the CODA literature is the roundingzero that is a component which is present but is too small to be detected bythe measurement instrument This is typical in chemical biological andgeological compositions and the CODA literature offers ample instruments todeal with rounding zeros

A classic essential zero example in economics is in household budget researchwhen measuring expenditure on tobacco and will essentially be zero if allmembers are non-smokers If tobacco expenditure is decidedly in the researchersrsquointerest the target population should be redefined to include only smokers Inmany instances budget research is often not clear whether zero expenditurescome closer to being essential or rounding zeros In some cases they can beunderstood as corner solutions in a utility maximization problem In others theycan be understood as the inherent randomness of human behaviour or as thelimitations of the data Tourists may spend a certain amount on activities oncertain trips but not on others and so surveys of only one trip will unavoidablycontain some zeros of this type Tourists may also forget or fail to report trivialexpenses such as postcard shopping local bus tickets and going to a museum(see Legoheacuterel (1998) for a discussion on the instability of tourism expenditure)Fry et al (2000) claim that in both situations zeros can be proxied by a verysmall value and thus be treated as rounding zeros In tourist budget researchto treat zero expenses in activities as rounding zeros implies assuming firstthat there is basically no tourist type who will never spend anything onactivities and second that tourists who generally spend little on activities arebasically similar to those who spend nothing or fail to report small expensesWe find both assumptions to be reasonable

Fry et al (2000) essentially used the same zero replacement strategy that waslater suggested by Martiacuten-Fernaacutendez et al (2003) namely replacing xid = 0 with

xprimeid = kδid with 0 lt k lt 1 (4)

where δid is the smallest detectable proportion for individual i and component d

17Determinants in tourist expenditure composition

Martiacuten-Fernaacutendez et al (2003) suggest using k = 065 although a sensitivityanalysis of the results on the choice of k is always advisable (we used k = 030and k = 099 with no sizeable change in the estimates) Next non-zero xid valueshave to be reduced in order to preserve the unit sum and the ratios among non-zero components As suggested by Martiacuten-Fernaacutendez et al (2003) with

xprimeid = xid (1 ndash Σxid=0

xprimeid) (5)

Simulations show this method performs particularly well if the proportion ofzeros is below 10 (Martiacuten-Fernaacutendez et al 2011)

In our data set zeros were present only in one budget category (discretionaryexpenditure) The minimum amount spent by the non-zero group was euro1which roughly corresponds to the price of a city bus ticket the entrance to asubsidized local museum or a cheap souvenir Since the total expenditureis known for each individual we compute δid by dividing euro1 with the totalexpenditure of individual I (see the appendix for the SPSS commandsyntax)

In this article we consider alternative log-ratio transformations which aremore flexible than the alr and clr in that the denominator does not have tobe the same in all ratios This increased flexibility makes it easier to computelog-ratios which are more interpretable with respect to the researchersrsquo questions orhypotheses

In general an interpretable log-ratio transformation is easy to computewhenever there is an interpretable sequential binary partition of componentsinto pairs of groups of components according to the researchersrsquo objectives orto the conceptual similarity of the components These partitions start bydividing components into two clusters and then continue by subdividingone of the clusters into two until each component constitutes its owncluster D components always involve Dndash1 partitions These partitions are bestunderstood as a partition tree or dendrogram (Pawlowsky-Glahn and Egozcue2011)

A meaningful log-ratio transformation takes ratios of the geometric meansof the two component clusters at each partition Numerators and denominatorsare interchangeable In our article we consider xi1 = transportation expenditurexi2 = accommodation and food (basic expenditure) and xi3 = activities andshopping (discretionary expenditure) An interpretable sequential partition isshown in Figure 1

The sequential partition in Figure 1 can have a dual interpretation It mayimply that researchers consider both types of at-destination expenses to bemutually similar and less similar to transportation Or it may imply that theresearch questions involve the distribution of total expenditure betweentransportation and at-destination expenditure and the distribution ofat-destination expenditure into basic and discretionary components as is thecase in our article

The first log-ratio compares transportation expenditure with the geometricmean of accommodation and food (basic expenditure) and activities andshopping (discretionary expenditure) With this ratio we want to observe howtravelling with an LCA or a legacy company affects the share of transportationcompared to non-transportation (at-destination) expenses

TOURISM ECONOMICS18

Figure 1 Sequential partition of total expenditure

⎛ xi1 ⎞ 1 1yi1 = ln ⎜ ndashndashndashndash ⎟ = ln(xi1) ndash ndash ln(xi2) ndash ndash ln(xi3) (6)

⎝ xi2xi3 ⎠ 2 2

Positive values show a transportation share greater than the geometric meanof the remaining two components Negative values show the opposite

The second log-ratio is a ratio of accommodation and food (basic expenditure)over activities and shopping (discretionary expenditure) With this ratio wewant to find out to what tourists allocate the rest of their trip budget andwhether more is allocated to basic or discretionary expenditure once they havepaid for (often in advance) their transportation

⎛ xi2 ⎞yi2 = ln ⎜ ndashndashndash ⎟ = ln(xi2) ndash ln(xi3) (7)

⎝ xi3 ⎠

Positive values show a basic (accommodation and food) share which is largerthan the discretionary (activities and shopping) share Negative values illustratethe opposite

Log-ratio transformations based on sequential binary partitions areproportional to the isometric log-ratio transformation (ilr see Egozcue et al2003 Egozcue and Pawlowsky-Glahn 2005)

As compositions are vector variables they cannot be analysed component-wise by means of univariate regression ANOVA models and the likeSeemingly unrelated regression models for continuous explanatory variables ormultivariate ANOVA (MANOVA) models for categorical explanatory variablesare appropriate MANOVArsquos multivariate tests and statistics (for example

x2 x3 Transportation

Global partition of total expenditure

x1

Accommodation

amp food

(basic)

Activities

amp shopping

(discretionary)

Partition of

expenses other

than transportation

(at-destination

expenses)

19Determinants in tourist expenditure composition

Pillairsquos trace Hotellingrsquos trace or Wilkrsquos lambda) are invariant to howcomponents are arranged in Figure 1 to compute log-ratios (Mateu-Figueras etal 2013) Univariate tests referring to each particular log-ratio are not invari-ant hence the importance of the interpretability of each log-ratio The softwareSPSS 19 is used to estimate the MANOVA model (GLM procedure)

We have to distinguish between three types of variables moderatingexogenous and endogenous to be included in the analysis

Moderating variables are those which modify the effect of exogenousvariables In our article we treat type of airline as a moderator in other wordswe include its interaction terms with all other variables in the MANOVA model

Exogenous variables are assumed to affect expenditure but not the converseSocio-demographic characteristics (travellerrsquos age gender education incomeetc) are obviously determined prior to any travel decision and hence exogenouswith respect to expenditure

Many of the choices tourists make are interdependent (Dellaert et al 1997)or at least planned simultaneously (Fesenmaier and Jeng 2000) So expenditureis arguably decided on at the same time as many trip attributes At least itis unclear whether expenditure is consistently decided on after trip attributesby all travellers We consider them to be endogenous variables and thereforedo not include them in the MANOVA model as explanatory in order to preventendogeneity problems Instead we display them graphically in the log-ratiospace The means of both log-ratios within each endogenous category are thecategory coordinates

The large sample size (see next subsection) makes it possible to use lowp-values Moderating effects with p-values higher than 001 according to eitherPillairsquos trace Hotellingrsquos trace or Wilksrsquo lambda were removed from the modelAll the variables and moderating effects included in the final model aresignificant at 001

Sample and variables

In this article we use secondary official statistics data The data were providedby the Instituto de Estudios Turiacutesticos (IET) an official agency of the Ministryof Industry Energy and Tourism and one which produces the majority oftourism data in Spain The survey is known as the Encuesta de Gasto Turiacutestico(EGATUR) in which tourism expenditure and other tourist information suchas trip information and tourist sociodemographic characteristics are studiedThe EGATUR survey conducted in 27 major Spanish airports in 2010 usedCAPI (computer assisted personal interview) to interview tourists leaving thecountry The sample is non-proportionally stratified by country of residenceairport and month See IET (2012) for further details on the EGATURmethodology

Our universe is a subset of the EGATUR universe which consists ofEuropean leisure visitors arriving by air and spending at least one night inSpain We excluded flights from outside Europe because LCAs mostly operateshort-haul flights We also centred our study on only those trips with one singledestination thus excluding multi-stage trips as the decision process regardingexpenditure composition for these trips is expected to fundamentally differ fromthat of single-stage trips Stays of over 120 days were also excluded

TOURISM ECONOMICS20

For this study we did not consider tourists

bull who have essential zeros in accommodation (tourists who own a house at thedestination or tourists who stay with friends or relatives)

bull who do not decide how much they will spend on certain components(business and study trips)

bull who do not pay for the trip themselves (trips paid for by employers familyfriends prizes offers etc)

bull for whom composition is wholly or partly unobserved (package tourists)

The final sample size was N= 19359From the expenditure variables included in the EGATUR survey database

and used in the model as budget components we first put the amount paidfor transportation (x1) This component has no zeros

Second the amount of money paid for accommodation and food isundistinguishable for full-board half-board or bed and breakfast accommodationTherefore we define a joint accommodation and food component (basic expendi-ture) In this component we include not only the amount paid for consumptionin bars and restaurants but also for buying groceries and everyday products insupermarkets (x2) This component has no zeros

Finally EGATUR provides an aggregated expenditure on activities andshopping (except groceries and everyday products) To this we added theconceptually similar amount paid for moving around the destination (publictransportation andor car rented at the destination) in order to build anactivities and shopping component (x3) This component had 98 zeros whichstresses its discretionary character Zeros were replaced as explained in thestatistical approach subsection

We consider as at-destination (or also called non-transportation) expenses thebasic component (food and accommodation) and the discretionary component(activities and shopping)

Share and log-ratios are described in Table 1 Explanatory variables are thelevel of education income country of residence gender travel groupprofessional status (for pensioners we consider their last professional position)and age The age variable is built as a combination of the original age variableand the variable referring to the individual economic situation This is donein order to have a specific pensioner category taking into account the varyingretirement ages across countries and professions and thus capturing thosetourists who have more free time to undertake a trip regardless of physical ageTable 2 shows their categories and frequencies

As endogenous variables which are not included in the MANOVA modelbut are included in the log-ratio space we include activities undertaken at thedestination accommodation length of stay total expenditure quartilesquartiles of expenditure made at destination per day (basic plus discretionaryexpenses) and a combination of motivation and destination This last variablehas been constructed as a combination of the motivation and the destinationvariable We distinguish between first those tourists travelling for culturaltourism to singular cities second those who come just for leisure either inthe countryside or more commonly at the seaside and third other typologies(Table 3)

21Determinants in tourist expenditure composition

Table 1 Percentage share and log-ratio descriptive statistics

Min Max Mean SD Skewness Kurtosis

Transportation component (x1) 003 9611 2846 1387 0837 0789Basic component (x2) 021 9963 5472 1613 ndash0114 ndash0450Discretionary component (x3) 001 8668 1681 1292 0988 1174Transportationat-destinationlog-ratio (y1) ndash757 548 01506 10702 0714 1404

Basicdiscretionary log-ratio (y2) ndash530 872 17657 18173 1573 2151

Results

Results for the exogenous variables

Table 4 presents the MANOVA results The univariate R2 corrected for thedegrees of freedom is 0045 (first log-ratio) and 0105 (second log-ratio) Themultivariate uncorrected R2 is 1 ndash Λ = 0178 and corrected for the degrees offreedom is 0176 We checked that the results did not change when removingthe four outliers with the highest Cookrsquos distance

Table 4 presents the estimates for the two log-ratios the log-ratio oftransportation over the other two categories (y1) and the log-ratio of basicexpenditure (accommodation and food) over discretionary expenditure (activitiesand shopping) (y2) respectively Because of the moderating effects within eachlog-ratio there are two columns representing the effects when flying with anLCA or with a legacy airline (main and moderating effects have already addedtogether for easier reading) If the LCA and legacy columns are different thereis a significant moderating effect (p-value lt 001)

As we are interpreting log-ratios a positive estimate of a given predictorcategory means that tourists in that category spend more on the numeratorcompared to the denominator than tourists in the reference predictor categoryA negative estimate shows the opposite

The intercept term shows the main effect of company type that is thepredicted log-ratio for each type of airline within the reference category for allvariables (university education medium income resident in the UK or Ireland25ndash44 yearsrsquo old male travelling with partner and mid-level employee) Withinthe reference categories legacy users spend a higher proportion on transportationcompared to other expenses and LCA users spend more on basic expensescompared to discretionary

The level of education seems to have more to do with activities undertakenthan with transportation Results show that a lower level of education resultsin higher expenditure on basic expenses compared to discretionary and is almostequal for users of both types of airline For mainly or only LCA users a lowerlevel of education results in a higher share in transport compared to the othertwo categories If we put it in another way there seems to be a distinct highlyeducated LCA user segment which utilizes the savings in transport toincrease expenditure in non-transportation expenses and even more so indiscretionary

TOURISM ECONOMICS22

Table 2 Frequency distributions of exogenous and moderating variables in the MANOVAmodel

Variables Count Percentage

Level of educationUp to high school 6019 311Universitya 13340 689

Income categoryMediumlow 797 41Mediuma 13186 681Mediumhigh 4376 226High 1000 52

Country of residencePortugal 624 32Other European countries 850 44Belgium 1006 52Netherlands 1069 55Austria Switzerland and Liechtenstein 1122 58France 1326 68Scandinavian countries 1789 92Germany 2363 122Italy 3113 181United Kingdom and Irelanda 6097 315

GenderFemale 8530 441Malea 10829 559

Travel groupAlone 2484 128In family 2980 154With friends 4440 229With partnera 9446 488

Professional statusHomemaker 517 27Unemployed 602 31Student 1432 74Low-level employee 915 47Mid-level employeea 10964 586High-level employee 2138 110Self-employed 2791 144

AgeOver 45 yearsrsquo old and pensioner 1484 77Over 45 yearsrsquo old and not pensioner 5009 25925ndash44 yearsrsquo olda 10569 54615ndash24 yearsrsquo old 2297 119

Type of airline (moderating variable)Legacy 5884 304Low cost 13475 696

Note aReference categories in the MANOVA model chosen either because they are the largestcategories or because they are the most standard conceptually considered

23Determinants in tourist expenditure composition

Table 3 Frequency distributions of endogenous variables (not included in the MANOVAmodel)

Variables Count Percentage

ActivitiesGolfYes 227 12No 19132 988HikingYes 364 19No 18995 981Sporting eventsYes 603 31No 18756 969Nautical sportsYes 1645 85No 17714 915Other sportsYes 904 47No 18455 953Cultural visitsYes 11180 578No 8179 422Cultural eventsYes 2452 127No 16907 873Other cultural activitiesYes 4175 216No 15184 784SpaYes 1199 62No 18160 938Theme parksYes 1745 90No 17614 910GastronomyYes 1565 81No 17794 919Type of accommodationHotel 4ndash5 5056 261Hotel 3 4922 254Hotel lt 3 6673 345Rented appartment 2116 109Other accommodation 592 31Length of stay5 or less nights 9911 5126ndash8 nights 5952 3079ndash12 nights 1592 8213ndash15 nights 1261 6516 and more nights 643 33

TOURISM ECONOMICS24

Table 3 continued

Variables Count Percentage

Total expenditureQuartile 1 5074 262Quartile 2 5104 264Quartile 3 4986 258Quartile 4 4195 217Expenditure at destination per dayQuartile 1 4855 251Quartile 2 4832 250Quartile 3 4868 251Quartile 4 4804 248Motivation-destinationLeisure on coastin countryside 8083 418Urban tourism 5324 275Others 5952 307

Level of income in contrast affects both log-ratios The higher the incomethe lower the transport share The higher the income the lower the basicdiscretionary expense ratio that is a higher portion of the non-transportexpense is devoted to the discretionary budget

As far as the country of residence is concerned the results show that it affectsmainly the basicdiscretionary log-ratio and exhibits strong airline typemoderating effects LCA users tend to differ depending on their country ofresidence when distributing non-transport expenses into basic versusdiscretionary In general when compared to the UK and Ireland residentcategory (reference category) all other European residents have a largerdiscretionary share More precisely the French Italians and Portuguese have thehighest discretionary share within LCA users and other European countries(mainly central and eastern European) have the highest discretionary sharewithin legacy airline users

Gender has a small effect which is constant for both types of airline Femaletravellers tend to slightly increase the transport share and the share of basicexpenses within the non-transport expenses

The travel group seems to affect both log-ratios and has a moderating effectwith airline type Those tourists who travel in a family group have a highershare in transport expenses when flying with legacy airlines In the second log-ratio travelling as a family with legacy airlines increases the basic expensesshare compared to the share in discretionary Travelling with friends increasesthe share in non-transport expenses for both airline types Besides those whotravel with friends spend more in the discretionary share compared to basicexpenses and those who fly with LCA even more so Finally travelling aloneincreases the basic expenses share compared to the share in discretionaryexpenses for both airline types

When professional status is considered the results show that it mostly affectsthe basicdiscretionary log-ratio and there are some relevant airline type

25Determinants in tourist expenditure composition

Table 4 MANOVA results

y1 log-ratio y2 log-ratiotransportation basicat-destination discretionary

LCA Legacy LCA Legacy

Intercept 0225 0274 2195 1881

Up to high school 0192 0078 0406 0437University 0 0 0 0

Mediumlow income 0073 0289 0037 0518Medium income 0 0 0 0Mediumhigh income ndash0227 ndash0058 ndash0269 ndash0006High income ndash0495 ndash0190 ndash0463 ndash0141

Residence in Portugal 0041 0069 ndash1136 ndash0580Residence in other European Countries 0326 ndash0012 ndash0772 ndash0820Residence in Belgium ndash0127 ndash0159 ndash0667 ndash0461Residence in Netherlands ndash0120 ndash0197 ndash0678 ndash0530Residence in Austria + Switzerland +Liechtenstein ndash0017 ndash0087 ndash0687 ndash0436

Residence in France ndash0112 ndash0111 ndash0870 ndash0553Residence in Scandinavian countries ndash0045 ndash0038 ndash0660 ndash0477Residence in Germany ndash0283 ndash0176 ndash0452 ndash0392Residence in Italy ndash0209 ndash0266 ndash0855 ndash0503Residence in UK or Ireland 0 0 0 0

Female 0094 0094 0093 0093Male 0 0 0 0

Travelling alone ndash0053 ndash0002 0081 0126Travelling in family 0085 0193 0014 0213Travelling with friends ndash0237 ndash0191 ndash0504 ndash0282Travelling with partner 0 0 0 0

Homemaker 0080 0160 0253 0235Unemployed 0030 0159 ndash0036 ndash0039Student ndash0005 ndash0104 ndash0331 ndash0624Low-level employee ndash0034 0194 ndash0194 0063Mid-level employee 0 0 0 0High-level employee ndash0008 ndash0064 ndash0001 ndash0139Self-employed 0029 ndash0086 0193 ndash0072

Over 45 yearsrsquo old and pensioner ndash0060 ndash0060 0633 0633Over 45 yearsrsquo old and not pensioner ndash0010 ndash0010 0142 014225ndash44 yearsrsquo old 0 0 0 015ndash24 yearsrsquo old 0002 0002 ndash0019 ndash0019

moderating effects Legacy users with a low professional status (homemakerunemployed or low-level employee) spend more on transport compared to thenon-transport expenses Conversely students using legacy airlines spend moreon non-transport expenses and these non-transport expenses are to a greater

TOURISM ECONOMICS26

Figure 2 Endogenous variables on the log-ratio space Mean log-ratio valueswithin each category of the endogenous variables

extent devoted to discretionary expenses compared to basic expensesHomemakers increase basic expenditure compared to discretionary with respectto the reference category (mid-level employee) for both airline types LCA userswho are self-employed also increase basic expenditure compared to discretionaryLow-level employees flying with an LCA increase the discretionary sharecompared to the basic and high-level employees flying with legacy airlines alsospend relatively more on the discretionary part

The results show that age affects the second log-ratio more than the firstand there is no moderating effect with the airline type Being a pensioner resultsin a higher share in basic expenses compared to discretionary expenses for bothairline types The remaining categories have hardly any differences from thereference 25ndash44 yearsrsquo old

Results for the endogenous variables

As introduced in the methodology section Figure 2 shows how the endogenousvariables behave in the log-ratio space The origin of the graph is representedby the mean log-ratios in Table 1 01506 for the transportother ratio and17657 for the basicdiscretionary ratio and according to this the graph showscategories with higher or lower ratios than the average

075

060

045

030

015

000

ndash015

ndash030

ndash045

ndash060

Ratio tra

nsport

ationa

t-destination e

xpenses

Ratio basicdiscretionary expenses

050 075 100 125 150 175 200 225 250

27Determinants in tourist expenditure composition

Regarding accommodation it seems that rented apartments and other typesof accommodation behave similarly and have the lowest budget share intransport In the case of hotels as the star rating increases expenditure on basiccompared to discretionary increases and expenditure on transport compared toat-destination decreases

As far as the length of stay is concerned longer stays increase the share ofnon-transport expenditure in basic expenses and reduce the share in transport

As regards destination and motivation tourists going to a seaside or country-side destination for leisure purposes behave quite differently from those goingto an urban destination for urban or cultural tourism The former group spendsmore on transport and less on the discretionary portion However those comingfor urban tourism show an increase in the share of non-transport expenses aswell as in the discretionary share compared to basic expenses

The total expenditure variable gives relevant information It is ordered alongthe first log-ratio axis from top to bottom meaning that those tourists havinga lower total expenditure spend relatively more on the transport share and thosewho have a higher total expenditure spend more on the other two parts of thebudget Regarding the expenditure made at destination per day it is ordereddiagonally from the upper-right corner to the bottom-left corner meaning thattourists with a higher at-destination expenditure per day spend less ontransportation and at the same time more on the discretionary part within at-destination expenses

Finally activities undertaken are located in the lower left quadrant of thegraph meaning that undertaking activities (discretionary expenses) decreasesthe share in transport expenses and although they continue to spend more onbasic expenses than on discretionary the discretionary share increases We havehighlighted some of them Those tourists attending sports and cultural eventsare those spending the most on discretionary compared to basic expensesHiking or some nautical sports could be free of charge thus tourists doing theseactivities have a basic to discretionary ratio close to the mean As for thetransportation share all activities tend to reduce this share about equally withthe exception of golf which leads to a much more substantial reduction

Conclusion and discussion

The main purpose of this article was to study the composition of touristexpenditure and its determinants that is the drivers of the share of touristexpenditure allocated to the different categories of a travel budget and withspecial emphasis on the distinction between legacy versus LCA travellers

The main differences between legacy and LCA users are that income somecountries of residence the travel group and some occupations have differenteffects depending on the type of airline Tourists with a medium or low incomewho travel with legacy airlines tend to spend a greater share of their travelbudget on transportation and basic expenses but those travelling with LCAstend to spend more at the destination and more specifically ondiscretionary expenditure LCA tourists residing in the lsquoother Europeancountriesrsquo spend more on transportation Families flying with legacy airlinesspend more on the transportation share compared to the at-destination share

TOURISM ECONOMICS28

and also spend more on basic expenses compared to discretionary ones Finallyin terms of occupational status legacy users with low-level employment spendmore on transportation whereas the opposite occurs with LCA travellers whospend more on the discretionary component Students and high-level employeesflying with legacy airlines spend more on the discretionary part of the tripbudget

As far as implications for management are concerned assuming that theobjectives of DMOs are to increase the allocation to the at-destinationcomponent of the trip budget (that is lowering the first log-ratio) some of therecommended actions would be to increase marketing efforts directed atmedium- and high-income earners especially at those travelling on LCAs toincrease marketing strategies in some of the European outbound markets suchas Germany (especially those using LCA) and Italy and the Netherlands(especially those flying with legacy airlines) DMOs should also focus on thosetourists travelling with friends (airline type is inconsequential) and studentsusing legacy companies As for the endogenous variables DMOs should directtheir efforts towards those tourists undertaking activities towards those stayinga little bit longer than a week and towards those with high total expenditureFurthermore if DMOs are also interested in capturing the groups who spendrelatively more on discretionary expenses (lowering the second log-ratio) theyshould focus their marketing efforts on those tourists who are highly educatedas well as those who have medium to high incomes and who use LCAsAdditionally they should focus on markets other than the UK and IrelandYoung tourists and tourists who travel with friends (where airline type isinconsequential) low-level employees and students using LCAs as well asstudents and high-level employees flying on legacy airlines should also betargeted In relation to the endogenous variables DMOs must take into accountthose tourists undertaking payable activities those on shorter stays or comingfor urban and cultural tourism or other purposes

The appeal of the CODA methodology for studying tourism budgets lies inthe fact that once the variables have been transformed the researcher can usestandard and well understood statistical models with unbounded error termdistributions while ensuring that the results will be compositionally coherentand that the standard statistical assumptions will hold Conversely results arenot compositionally coherent when we use the MANOVA model on raw sharewithout transforming it into log-ratios We have seen with our data that 84of individuals included in the sample have at least one limit of the 90 shareprediction intervals outside the [01] range

Contrary to the classic and highly restrictive alr and clr transformations theCODA methodology offers the potential to construct tailor-made log-ratios whichare intuitive to interpret and suit the research questions at hand A classificationtree of components is a clear and useful tool in this respect We encourageresearchers to use this approach in further research on tourism expenditure whenbudget share division is fundamental to the researchersrsquo questions

In this study we have encountered two limitations First contrary to studiesthat use macro data in micro data studies the coexistence of full-board halfboard bed and breakfast and accommodation-only tourists makes it impossibleto separate accommodation and food expenditure meaningfully Othersubdivisions (for example discretionary expenditure on activities non-grocery

29Determinants in tourist expenditure composition

shopping and moving around the destination) are technically feasible but canonly increase the percentages of zeros The CODA methodology involves somedegree of amalgamation of components Second even though there are someadvantages to using a database from an official statistics institution as in thiscase (large sample size and scope) the main disadvantage is that the set ofavailable variables cannot be controlled by the researcher A further issue to beaddressed is the time dimension As the EGATUR survey is conducted annuallyfurther research can be done to include a repeat cross-section analysis in orderto capture trends in the effects of predictors

References

Aitchison J (1986) The Statistical Analysis of Compositional Data Monographs on Statistics and AppliedProbability Chapman and Hall London (Reprinted in 2003 with additional material by TheBlackburn Press)

Alegre J Mateo S and Pou LL (2009) lsquoParticipation in tourism consumption and the intensityof participation an analysis of their socio-demographic and economic determinantsrsquo TourismEconomics Vol 15 No 3 pp 531ndash546

Bewley R (1982) lsquoThe generalized addilog demand system applied to Australian time series andcross section datarsquo Australian Economic Papers Vol 21 No 38 pp 177ndash192

Bewley R and Fiebig DG (1988) lsquoEstimation of price elasticities for an international telephonedemand modelrsquo The Journal of Industrial Economics Vol 36 No 4 pp 393ndash409

Cai LA (1998) lsquoAnalysing household food expenditure patterns on trips and vacations a Tobitmodersquo Journal of Hospitality and Tourism Research Vol 22 No 4 pp 338ndash358

Cai LA (1999) lsquoRelationship of household characteristics and lodging expenditure on leisure tripsrsquoJournal of Hospitality and Leisure Marketing Vol 6 No 2 pp 5ndash18

Cai LA Hong G-S and Morrison AM (1995) lsquoHousehold expenditure patterns for tourismproducts and servicesrsquo Journal of Travel and Tourism Marketing Vol 4 No 4 pp 15ndash39

Cannon TF and Ford J (2002) lsquoRelationship of demographic and trip characteristics to visitorspending an analysis of sports travel visitors across timersquo Tourism Economics Vol 8 No 3 pp 263ndash271

Castillo-Manzano JL and Marchena-Goacutemez M (2010) lsquoAnalysis of determinants of airline choiceprofiling the LCC passengerrsquo Applied Economics Letters Vol 18 No 1 pp 49ndash53

Chiou Y-C and Chen Y-H (2010) lsquoFactors influencing the intentions of passengers regarding full ser-vice and low cost carriers a notersquo Journal of Air Transport Management Vol 16 No 4 pp 226ndash228

Coenen M and van Eekeren L (2003) lsquoA study of the demand for domestic tourism by Swedishhouseholds using a two-staged budgeting modelrsquo Scandinavian Journal of Hospitality and TourismVol 3 No 2 pp 114ndash133

Craggs R and Schofield P (2009) lsquoExpenditure-based segmentation and visitor profiling at theQuays in Salford UKrsquo Tourism Economics Vol 15 No 1 pp 243ndash260

Deaton A and Muellbauer J (1980) lsquoAn almost ideal demand systemrsquo American Economic ReviewVol 70 No 3 pp 312ndash326

Dellaert B Borgers A and Timmermans H (1997) lsquoConsumer activity pattern choice devel-opment and test of stage-dependent conjoint choice experimentsrsquo Journal of Retailing and ConsumerServices Vol 4 No 1 pp 25ndash37

Divisekera S (2007) Modelling and Estimation of Tourism Demand Elasticities A Study of TourismExpenditure Allocation in Australia Cooperative Research Centre for Sustainable Tourism (CRCST)Gold Coast

Divisekera S (2009) lsquoEx-post demand for Australian tourism goods and servicesrsquo Tourism EconomicsVol 15 No 1 pp 153ndash180

Divisekera S (2010) lsquoEconomics of leisure and non-leisure tourist demand a study of domesticdemand for Australian tourismrsquo Tourism Economics Vol 16 No 1 pp 117ndash136

Dolnicar S Crouch GI Devinney T Huybers T Louviere JJ and Oppewal H (2008)lsquoTourism and discretionary income allocation Heterogeneity among householdsrsquo TourismManagement Vol 29 No 1 pp 44ndash52

Downward P and Lumsdon L (2000) lsquoThe demand for day-visits an analysis of visitor spendingrsquoTourism Economics Vol 6 No 3 pp 251ndash261

15Determinants in tourist expenditure composition

The fact that the error term in any proper system of demand equationsapplied to budget share should take into account the data compositional naturehas been widely acknowledged (Aitchison 1986 Ronning 1992 Fry et al1996) However the review in Fry (2011) reports few studies that have appliedthe CODA methodology to demand equations and to the study of householdbudgets (McLaren et al 1995 Fry et al 1996 2000)

When fitting demand equations to log-ratios of expenditure componentsEngelrsquos aggregation condition which requires that household total expenditureshould equal the sum of components is automatically satisfied The model caneasily be extended to a full consumer demand analysis by the introduction ofprices and other covariates such as consumer characteristics (Aitchison 1986)McLaren et al (1995) relate the CODA analysis with log-ratio transformationto the almost ideal demand system and conclude that CODA makes it possibleto reach the same objectives with normal and homoscedastic error terms

Related developments are the indirect addilog system (Houthakker 1960)and the generalized addilog system (Bewley 1982) When applied tocompositions (for example Bewley and Fiebig 1988) the latter is equivalentto the CODA methodology in which the log-ratios of each component over thegeometric means of all components are the dependent variables in the set ofsimultaneous regressions

To the best of our knowledge there is no study of tourism budget share usingthe CODA methodology or any other methodology accounting for thecompositional constraints in compositional data

Method and data

Statistical approach

The simplest CODA approach involves applying standard statistical techniqueson logarithms of ratios of components Several log-ratio transformations havebeen suggested in the early CODA literature (Egozcue et al 2003) The additivelog-ratio transformation (alr) used by Fry et al (1996 2000) is the most popularand the easiest to compute given that it is simply the log-ratio of eachcomponent to the last

yid = ln(xidxiD) = ln(xid) ndash ln(xiD) with d = 123D ndash 1 (2)

The centred log-ratio transformation (clr) used by the generalized addilogdemand system (Bewley 1982) computes the log-ratios of each component overthe geometric mean of all the components including itself

⎛ xid ⎞yid = ln ⎜ ndashndashndashndashndashndashndashndashndashndashndashndash ⎟ with d = 123D (3)

⎝ D xi1xi2xi3 xiD ⎠

All log-ratio transformed yid variables recover the full unconstrained ndashinfin to infinrange It must be noted that one dimension is lost in the alr while in the clrone dimension is a linear combination of the remaining

TOURISM ECONOMICS16

The alr is commonly used for statistical modelling and prediction ofcompositions (Fry et al 1996) Conversely the clr transformation is commonlyused for statistical techniques which are based on a metric such as a clusteranalysis because of its preservation of distances even though it leads to asingular covariance matrix Thus while the alr would be appropriate for thepurpose of this article the fact that one component must be used as referencefor all others reduces its flexibility and interpretability Alternatives arepresented at the end of this subsection

While having zero expenditures in absolute data indeed has significantmethodological consequences (Lee 2001) these consequences are arguably moreserious in the CODA methodology If the xid variables contain zeros then log-ratios cannot be computed An obvious initial procedure to reduce zeros is toamalgamate small and conceptually similar components with many zeros intolarger ones In tourism budget research it can be useful to group together allexpenditure on activities or all expenditure on food for instance

In certain instances some zero components result from individual character-istics which are called essential zeros in the CODA literature (Aitchison 1986)Another typology of zeros encountered in the CODA literature is the roundingzero that is a component which is present but is too small to be detected bythe measurement instrument This is typical in chemical biological andgeological compositions and the CODA literature offers ample instruments todeal with rounding zeros

A classic essential zero example in economics is in household budget researchwhen measuring expenditure on tobacco and will essentially be zero if allmembers are non-smokers If tobacco expenditure is decidedly in the researchersrsquointerest the target population should be redefined to include only smokers Inmany instances budget research is often not clear whether zero expenditurescome closer to being essential or rounding zeros In some cases they can beunderstood as corner solutions in a utility maximization problem In others theycan be understood as the inherent randomness of human behaviour or as thelimitations of the data Tourists may spend a certain amount on activities oncertain trips but not on others and so surveys of only one trip will unavoidablycontain some zeros of this type Tourists may also forget or fail to report trivialexpenses such as postcard shopping local bus tickets and going to a museum(see Legoheacuterel (1998) for a discussion on the instability of tourism expenditure)Fry et al (2000) claim that in both situations zeros can be proxied by a verysmall value and thus be treated as rounding zeros In tourist budget researchto treat zero expenses in activities as rounding zeros implies assuming firstthat there is basically no tourist type who will never spend anything onactivities and second that tourists who generally spend little on activities arebasically similar to those who spend nothing or fail to report small expensesWe find both assumptions to be reasonable

Fry et al (2000) essentially used the same zero replacement strategy that waslater suggested by Martiacuten-Fernaacutendez et al (2003) namely replacing xid = 0 with

xprimeid = kδid with 0 lt k lt 1 (4)

where δid is the smallest detectable proportion for individual i and component d

17Determinants in tourist expenditure composition

Martiacuten-Fernaacutendez et al (2003) suggest using k = 065 although a sensitivityanalysis of the results on the choice of k is always advisable (we used k = 030and k = 099 with no sizeable change in the estimates) Next non-zero xid valueshave to be reduced in order to preserve the unit sum and the ratios among non-zero components As suggested by Martiacuten-Fernaacutendez et al (2003) with

xprimeid = xid (1 ndash Σxid=0

xprimeid) (5)

Simulations show this method performs particularly well if the proportion ofzeros is below 10 (Martiacuten-Fernaacutendez et al 2011)

In our data set zeros were present only in one budget category (discretionaryexpenditure) The minimum amount spent by the non-zero group was euro1which roughly corresponds to the price of a city bus ticket the entrance to asubsidized local museum or a cheap souvenir Since the total expenditureis known for each individual we compute δid by dividing euro1 with the totalexpenditure of individual I (see the appendix for the SPSS commandsyntax)

In this article we consider alternative log-ratio transformations which aremore flexible than the alr and clr in that the denominator does not have tobe the same in all ratios This increased flexibility makes it easier to computelog-ratios which are more interpretable with respect to the researchersrsquo questions orhypotheses

In general an interpretable log-ratio transformation is easy to computewhenever there is an interpretable sequential binary partition of componentsinto pairs of groups of components according to the researchersrsquo objectives orto the conceptual similarity of the components These partitions start bydividing components into two clusters and then continue by subdividingone of the clusters into two until each component constitutes its owncluster D components always involve Dndash1 partitions These partitions are bestunderstood as a partition tree or dendrogram (Pawlowsky-Glahn and Egozcue2011)

A meaningful log-ratio transformation takes ratios of the geometric meansof the two component clusters at each partition Numerators and denominatorsare interchangeable In our article we consider xi1 = transportation expenditurexi2 = accommodation and food (basic expenditure) and xi3 = activities andshopping (discretionary expenditure) An interpretable sequential partition isshown in Figure 1

The sequential partition in Figure 1 can have a dual interpretation It mayimply that researchers consider both types of at-destination expenses to bemutually similar and less similar to transportation Or it may imply that theresearch questions involve the distribution of total expenditure betweentransportation and at-destination expenditure and the distribution ofat-destination expenditure into basic and discretionary components as is thecase in our article

The first log-ratio compares transportation expenditure with the geometricmean of accommodation and food (basic expenditure) and activities andshopping (discretionary expenditure) With this ratio we want to observe howtravelling with an LCA or a legacy company affects the share of transportationcompared to non-transportation (at-destination) expenses

TOURISM ECONOMICS18

Figure 1 Sequential partition of total expenditure

⎛ xi1 ⎞ 1 1yi1 = ln ⎜ ndashndashndashndash ⎟ = ln(xi1) ndash ndash ln(xi2) ndash ndash ln(xi3) (6)

⎝ xi2xi3 ⎠ 2 2

Positive values show a transportation share greater than the geometric meanof the remaining two components Negative values show the opposite

The second log-ratio is a ratio of accommodation and food (basic expenditure)over activities and shopping (discretionary expenditure) With this ratio wewant to find out to what tourists allocate the rest of their trip budget andwhether more is allocated to basic or discretionary expenditure once they havepaid for (often in advance) their transportation

⎛ xi2 ⎞yi2 = ln ⎜ ndashndashndash ⎟ = ln(xi2) ndash ln(xi3) (7)

⎝ xi3 ⎠

Positive values show a basic (accommodation and food) share which is largerthan the discretionary (activities and shopping) share Negative values illustratethe opposite

Log-ratio transformations based on sequential binary partitions areproportional to the isometric log-ratio transformation (ilr see Egozcue et al2003 Egozcue and Pawlowsky-Glahn 2005)

As compositions are vector variables they cannot be analysed component-wise by means of univariate regression ANOVA models and the likeSeemingly unrelated regression models for continuous explanatory variables ormultivariate ANOVA (MANOVA) models for categorical explanatory variablesare appropriate MANOVArsquos multivariate tests and statistics (for example

x2 x3 Transportation

Global partition of total expenditure

x1

Accommodation

amp food

(basic)

Activities

amp shopping

(discretionary)

Partition of

expenses other

than transportation

(at-destination

expenses)

19Determinants in tourist expenditure composition

Pillairsquos trace Hotellingrsquos trace or Wilkrsquos lambda) are invariant to howcomponents are arranged in Figure 1 to compute log-ratios (Mateu-Figueras etal 2013) Univariate tests referring to each particular log-ratio are not invari-ant hence the importance of the interpretability of each log-ratio The softwareSPSS 19 is used to estimate the MANOVA model (GLM procedure)

We have to distinguish between three types of variables moderatingexogenous and endogenous to be included in the analysis

Moderating variables are those which modify the effect of exogenousvariables In our article we treat type of airline as a moderator in other wordswe include its interaction terms with all other variables in the MANOVA model

Exogenous variables are assumed to affect expenditure but not the converseSocio-demographic characteristics (travellerrsquos age gender education incomeetc) are obviously determined prior to any travel decision and hence exogenouswith respect to expenditure

Many of the choices tourists make are interdependent (Dellaert et al 1997)or at least planned simultaneously (Fesenmaier and Jeng 2000) So expenditureis arguably decided on at the same time as many trip attributes At least itis unclear whether expenditure is consistently decided on after trip attributesby all travellers We consider them to be endogenous variables and thereforedo not include them in the MANOVA model as explanatory in order to preventendogeneity problems Instead we display them graphically in the log-ratiospace The means of both log-ratios within each endogenous category are thecategory coordinates

The large sample size (see next subsection) makes it possible to use lowp-values Moderating effects with p-values higher than 001 according to eitherPillairsquos trace Hotellingrsquos trace or Wilksrsquo lambda were removed from the modelAll the variables and moderating effects included in the final model aresignificant at 001

Sample and variables

In this article we use secondary official statistics data The data were providedby the Instituto de Estudios Turiacutesticos (IET) an official agency of the Ministryof Industry Energy and Tourism and one which produces the majority oftourism data in Spain The survey is known as the Encuesta de Gasto Turiacutestico(EGATUR) in which tourism expenditure and other tourist information suchas trip information and tourist sociodemographic characteristics are studiedThe EGATUR survey conducted in 27 major Spanish airports in 2010 usedCAPI (computer assisted personal interview) to interview tourists leaving thecountry The sample is non-proportionally stratified by country of residenceairport and month See IET (2012) for further details on the EGATURmethodology

Our universe is a subset of the EGATUR universe which consists ofEuropean leisure visitors arriving by air and spending at least one night inSpain We excluded flights from outside Europe because LCAs mostly operateshort-haul flights We also centred our study on only those trips with one singledestination thus excluding multi-stage trips as the decision process regardingexpenditure composition for these trips is expected to fundamentally differ fromthat of single-stage trips Stays of over 120 days were also excluded

TOURISM ECONOMICS20

For this study we did not consider tourists

bull who have essential zeros in accommodation (tourists who own a house at thedestination or tourists who stay with friends or relatives)

bull who do not decide how much they will spend on certain components(business and study trips)

bull who do not pay for the trip themselves (trips paid for by employers familyfriends prizes offers etc)

bull for whom composition is wholly or partly unobserved (package tourists)

The final sample size was N= 19359From the expenditure variables included in the EGATUR survey database

and used in the model as budget components we first put the amount paidfor transportation (x1) This component has no zeros

Second the amount of money paid for accommodation and food isundistinguishable for full-board half-board or bed and breakfast accommodationTherefore we define a joint accommodation and food component (basic expendi-ture) In this component we include not only the amount paid for consumptionin bars and restaurants but also for buying groceries and everyday products insupermarkets (x2) This component has no zeros

Finally EGATUR provides an aggregated expenditure on activities andshopping (except groceries and everyday products) To this we added theconceptually similar amount paid for moving around the destination (publictransportation andor car rented at the destination) in order to build anactivities and shopping component (x3) This component had 98 zeros whichstresses its discretionary character Zeros were replaced as explained in thestatistical approach subsection

We consider as at-destination (or also called non-transportation) expenses thebasic component (food and accommodation) and the discretionary component(activities and shopping)

Share and log-ratios are described in Table 1 Explanatory variables are thelevel of education income country of residence gender travel groupprofessional status (for pensioners we consider their last professional position)and age The age variable is built as a combination of the original age variableand the variable referring to the individual economic situation This is donein order to have a specific pensioner category taking into account the varyingretirement ages across countries and professions and thus capturing thosetourists who have more free time to undertake a trip regardless of physical ageTable 2 shows their categories and frequencies

As endogenous variables which are not included in the MANOVA modelbut are included in the log-ratio space we include activities undertaken at thedestination accommodation length of stay total expenditure quartilesquartiles of expenditure made at destination per day (basic plus discretionaryexpenses) and a combination of motivation and destination This last variablehas been constructed as a combination of the motivation and the destinationvariable We distinguish between first those tourists travelling for culturaltourism to singular cities second those who come just for leisure either inthe countryside or more commonly at the seaside and third other typologies(Table 3)

21Determinants in tourist expenditure composition

Table 1 Percentage share and log-ratio descriptive statistics

Min Max Mean SD Skewness Kurtosis

Transportation component (x1) 003 9611 2846 1387 0837 0789Basic component (x2) 021 9963 5472 1613 ndash0114 ndash0450Discretionary component (x3) 001 8668 1681 1292 0988 1174Transportationat-destinationlog-ratio (y1) ndash757 548 01506 10702 0714 1404

Basicdiscretionary log-ratio (y2) ndash530 872 17657 18173 1573 2151

Results

Results for the exogenous variables

Table 4 presents the MANOVA results The univariate R2 corrected for thedegrees of freedom is 0045 (first log-ratio) and 0105 (second log-ratio) Themultivariate uncorrected R2 is 1 ndash Λ = 0178 and corrected for the degrees offreedom is 0176 We checked that the results did not change when removingthe four outliers with the highest Cookrsquos distance

Table 4 presents the estimates for the two log-ratios the log-ratio oftransportation over the other two categories (y1) and the log-ratio of basicexpenditure (accommodation and food) over discretionary expenditure (activitiesand shopping) (y2) respectively Because of the moderating effects within eachlog-ratio there are two columns representing the effects when flying with anLCA or with a legacy airline (main and moderating effects have already addedtogether for easier reading) If the LCA and legacy columns are different thereis a significant moderating effect (p-value lt 001)

As we are interpreting log-ratios a positive estimate of a given predictorcategory means that tourists in that category spend more on the numeratorcompared to the denominator than tourists in the reference predictor categoryA negative estimate shows the opposite

The intercept term shows the main effect of company type that is thepredicted log-ratio for each type of airline within the reference category for allvariables (university education medium income resident in the UK or Ireland25ndash44 yearsrsquo old male travelling with partner and mid-level employee) Withinthe reference categories legacy users spend a higher proportion on transportationcompared to other expenses and LCA users spend more on basic expensescompared to discretionary

The level of education seems to have more to do with activities undertakenthan with transportation Results show that a lower level of education resultsin higher expenditure on basic expenses compared to discretionary and is almostequal for users of both types of airline For mainly or only LCA users a lowerlevel of education results in a higher share in transport compared to the othertwo categories If we put it in another way there seems to be a distinct highlyeducated LCA user segment which utilizes the savings in transport toincrease expenditure in non-transportation expenses and even more so indiscretionary

TOURISM ECONOMICS22

Table 2 Frequency distributions of exogenous and moderating variables in the MANOVAmodel

Variables Count Percentage

Level of educationUp to high school 6019 311Universitya 13340 689

Income categoryMediumlow 797 41Mediuma 13186 681Mediumhigh 4376 226High 1000 52

Country of residencePortugal 624 32Other European countries 850 44Belgium 1006 52Netherlands 1069 55Austria Switzerland and Liechtenstein 1122 58France 1326 68Scandinavian countries 1789 92Germany 2363 122Italy 3113 181United Kingdom and Irelanda 6097 315

GenderFemale 8530 441Malea 10829 559

Travel groupAlone 2484 128In family 2980 154With friends 4440 229With partnera 9446 488

Professional statusHomemaker 517 27Unemployed 602 31Student 1432 74Low-level employee 915 47Mid-level employeea 10964 586High-level employee 2138 110Self-employed 2791 144

AgeOver 45 yearsrsquo old and pensioner 1484 77Over 45 yearsrsquo old and not pensioner 5009 25925ndash44 yearsrsquo olda 10569 54615ndash24 yearsrsquo old 2297 119

Type of airline (moderating variable)Legacy 5884 304Low cost 13475 696

Note aReference categories in the MANOVA model chosen either because they are the largestcategories or because they are the most standard conceptually considered

23Determinants in tourist expenditure composition

Table 3 Frequency distributions of endogenous variables (not included in the MANOVAmodel)

Variables Count Percentage

ActivitiesGolfYes 227 12No 19132 988HikingYes 364 19No 18995 981Sporting eventsYes 603 31No 18756 969Nautical sportsYes 1645 85No 17714 915Other sportsYes 904 47No 18455 953Cultural visitsYes 11180 578No 8179 422Cultural eventsYes 2452 127No 16907 873Other cultural activitiesYes 4175 216No 15184 784SpaYes 1199 62No 18160 938Theme parksYes 1745 90No 17614 910GastronomyYes 1565 81No 17794 919Type of accommodationHotel 4ndash5 5056 261Hotel 3 4922 254Hotel lt 3 6673 345Rented appartment 2116 109Other accommodation 592 31Length of stay5 or less nights 9911 5126ndash8 nights 5952 3079ndash12 nights 1592 8213ndash15 nights 1261 6516 and more nights 643 33

TOURISM ECONOMICS24

Table 3 continued

Variables Count Percentage

Total expenditureQuartile 1 5074 262Quartile 2 5104 264Quartile 3 4986 258Quartile 4 4195 217Expenditure at destination per dayQuartile 1 4855 251Quartile 2 4832 250Quartile 3 4868 251Quartile 4 4804 248Motivation-destinationLeisure on coastin countryside 8083 418Urban tourism 5324 275Others 5952 307

Level of income in contrast affects both log-ratios The higher the incomethe lower the transport share The higher the income the lower the basicdiscretionary expense ratio that is a higher portion of the non-transportexpense is devoted to the discretionary budget

As far as the country of residence is concerned the results show that it affectsmainly the basicdiscretionary log-ratio and exhibits strong airline typemoderating effects LCA users tend to differ depending on their country ofresidence when distributing non-transport expenses into basic versusdiscretionary In general when compared to the UK and Ireland residentcategory (reference category) all other European residents have a largerdiscretionary share More precisely the French Italians and Portuguese have thehighest discretionary share within LCA users and other European countries(mainly central and eastern European) have the highest discretionary sharewithin legacy airline users

Gender has a small effect which is constant for both types of airline Femaletravellers tend to slightly increase the transport share and the share of basicexpenses within the non-transport expenses

The travel group seems to affect both log-ratios and has a moderating effectwith airline type Those tourists who travel in a family group have a highershare in transport expenses when flying with legacy airlines In the second log-ratio travelling as a family with legacy airlines increases the basic expensesshare compared to the share in discretionary Travelling with friends increasesthe share in non-transport expenses for both airline types Besides those whotravel with friends spend more in the discretionary share compared to basicexpenses and those who fly with LCA even more so Finally travelling aloneincreases the basic expenses share compared to the share in discretionaryexpenses for both airline types

When professional status is considered the results show that it mostly affectsthe basicdiscretionary log-ratio and there are some relevant airline type

25Determinants in tourist expenditure composition

Table 4 MANOVA results

y1 log-ratio y2 log-ratiotransportation basicat-destination discretionary

LCA Legacy LCA Legacy

Intercept 0225 0274 2195 1881

Up to high school 0192 0078 0406 0437University 0 0 0 0

Mediumlow income 0073 0289 0037 0518Medium income 0 0 0 0Mediumhigh income ndash0227 ndash0058 ndash0269 ndash0006High income ndash0495 ndash0190 ndash0463 ndash0141

Residence in Portugal 0041 0069 ndash1136 ndash0580Residence in other European Countries 0326 ndash0012 ndash0772 ndash0820Residence in Belgium ndash0127 ndash0159 ndash0667 ndash0461Residence in Netherlands ndash0120 ndash0197 ndash0678 ndash0530Residence in Austria + Switzerland +Liechtenstein ndash0017 ndash0087 ndash0687 ndash0436

Residence in France ndash0112 ndash0111 ndash0870 ndash0553Residence in Scandinavian countries ndash0045 ndash0038 ndash0660 ndash0477Residence in Germany ndash0283 ndash0176 ndash0452 ndash0392Residence in Italy ndash0209 ndash0266 ndash0855 ndash0503Residence in UK or Ireland 0 0 0 0

Female 0094 0094 0093 0093Male 0 0 0 0

Travelling alone ndash0053 ndash0002 0081 0126Travelling in family 0085 0193 0014 0213Travelling with friends ndash0237 ndash0191 ndash0504 ndash0282Travelling with partner 0 0 0 0

Homemaker 0080 0160 0253 0235Unemployed 0030 0159 ndash0036 ndash0039Student ndash0005 ndash0104 ndash0331 ndash0624Low-level employee ndash0034 0194 ndash0194 0063Mid-level employee 0 0 0 0High-level employee ndash0008 ndash0064 ndash0001 ndash0139Self-employed 0029 ndash0086 0193 ndash0072

Over 45 yearsrsquo old and pensioner ndash0060 ndash0060 0633 0633Over 45 yearsrsquo old and not pensioner ndash0010 ndash0010 0142 014225ndash44 yearsrsquo old 0 0 0 015ndash24 yearsrsquo old 0002 0002 ndash0019 ndash0019

moderating effects Legacy users with a low professional status (homemakerunemployed or low-level employee) spend more on transport compared to thenon-transport expenses Conversely students using legacy airlines spend moreon non-transport expenses and these non-transport expenses are to a greater

TOURISM ECONOMICS26

Figure 2 Endogenous variables on the log-ratio space Mean log-ratio valueswithin each category of the endogenous variables

extent devoted to discretionary expenses compared to basic expensesHomemakers increase basic expenditure compared to discretionary with respectto the reference category (mid-level employee) for both airline types LCA userswho are self-employed also increase basic expenditure compared to discretionaryLow-level employees flying with an LCA increase the discretionary sharecompared to the basic and high-level employees flying with legacy airlines alsospend relatively more on the discretionary part

The results show that age affects the second log-ratio more than the firstand there is no moderating effect with the airline type Being a pensioner resultsin a higher share in basic expenses compared to discretionary expenses for bothairline types The remaining categories have hardly any differences from thereference 25ndash44 yearsrsquo old

Results for the endogenous variables

As introduced in the methodology section Figure 2 shows how the endogenousvariables behave in the log-ratio space The origin of the graph is representedby the mean log-ratios in Table 1 01506 for the transportother ratio and17657 for the basicdiscretionary ratio and according to this the graph showscategories with higher or lower ratios than the average

075

060

045

030

015

000

ndash015

ndash030

ndash045

ndash060

Ratio tra

nsport

ationa

t-destination e

xpenses

Ratio basicdiscretionary expenses

050 075 100 125 150 175 200 225 250

27Determinants in tourist expenditure composition

Regarding accommodation it seems that rented apartments and other typesof accommodation behave similarly and have the lowest budget share intransport In the case of hotels as the star rating increases expenditure on basiccompared to discretionary increases and expenditure on transport compared toat-destination decreases

As far as the length of stay is concerned longer stays increase the share ofnon-transport expenditure in basic expenses and reduce the share in transport

As regards destination and motivation tourists going to a seaside or country-side destination for leisure purposes behave quite differently from those goingto an urban destination for urban or cultural tourism The former group spendsmore on transport and less on the discretionary portion However those comingfor urban tourism show an increase in the share of non-transport expenses aswell as in the discretionary share compared to basic expenses

The total expenditure variable gives relevant information It is ordered alongthe first log-ratio axis from top to bottom meaning that those tourists havinga lower total expenditure spend relatively more on the transport share and thosewho have a higher total expenditure spend more on the other two parts of thebudget Regarding the expenditure made at destination per day it is ordereddiagonally from the upper-right corner to the bottom-left corner meaning thattourists with a higher at-destination expenditure per day spend less ontransportation and at the same time more on the discretionary part within at-destination expenses

Finally activities undertaken are located in the lower left quadrant of thegraph meaning that undertaking activities (discretionary expenses) decreasesthe share in transport expenses and although they continue to spend more onbasic expenses than on discretionary the discretionary share increases We havehighlighted some of them Those tourists attending sports and cultural eventsare those spending the most on discretionary compared to basic expensesHiking or some nautical sports could be free of charge thus tourists doing theseactivities have a basic to discretionary ratio close to the mean As for thetransportation share all activities tend to reduce this share about equally withthe exception of golf which leads to a much more substantial reduction

Conclusion and discussion

The main purpose of this article was to study the composition of touristexpenditure and its determinants that is the drivers of the share of touristexpenditure allocated to the different categories of a travel budget and withspecial emphasis on the distinction between legacy versus LCA travellers

The main differences between legacy and LCA users are that income somecountries of residence the travel group and some occupations have differenteffects depending on the type of airline Tourists with a medium or low incomewho travel with legacy airlines tend to spend a greater share of their travelbudget on transportation and basic expenses but those travelling with LCAstend to spend more at the destination and more specifically ondiscretionary expenditure LCA tourists residing in the lsquoother Europeancountriesrsquo spend more on transportation Families flying with legacy airlinesspend more on the transportation share compared to the at-destination share

TOURISM ECONOMICS28

and also spend more on basic expenses compared to discretionary ones Finallyin terms of occupational status legacy users with low-level employment spendmore on transportation whereas the opposite occurs with LCA travellers whospend more on the discretionary component Students and high-level employeesflying with legacy airlines spend more on the discretionary part of the tripbudget

As far as implications for management are concerned assuming that theobjectives of DMOs are to increase the allocation to the at-destinationcomponent of the trip budget (that is lowering the first log-ratio) some of therecommended actions would be to increase marketing efforts directed atmedium- and high-income earners especially at those travelling on LCAs toincrease marketing strategies in some of the European outbound markets suchas Germany (especially those using LCA) and Italy and the Netherlands(especially those flying with legacy airlines) DMOs should also focus on thosetourists travelling with friends (airline type is inconsequential) and studentsusing legacy companies As for the endogenous variables DMOs should directtheir efforts towards those tourists undertaking activities towards those stayinga little bit longer than a week and towards those with high total expenditureFurthermore if DMOs are also interested in capturing the groups who spendrelatively more on discretionary expenses (lowering the second log-ratio) theyshould focus their marketing efforts on those tourists who are highly educatedas well as those who have medium to high incomes and who use LCAsAdditionally they should focus on markets other than the UK and IrelandYoung tourists and tourists who travel with friends (where airline type isinconsequential) low-level employees and students using LCAs as well asstudents and high-level employees flying on legacy airlines should also betargeted In relation to the endogenous variables DMOs must take into accountthose tourists undertaking payable activities those on shorter stays or comingfor urban and cultural tourism or other purposes

The appeal of the CODA methodology for studying tourism budgets lies inthe fact that once the variables have been transformed the researcher can usestandard and well understood statistical models with unbounded error termdistributions while ensuring that the results will be compositionally coherentand that the standard statistical assumptions will hold Conversely results arenot compositionally coherent when we use the MANOVA model on raw sharewithout transforming it into log-ratios We have seen with our data that 84of individuals included in the sample have at least one limit of the 90 shareprediction intervals outside the [01] range

Contrary to the classic and highly restrictive alr and clr transformations theCODA methodology offers the potential to construct tailor-made log-ratios whichare intuitive to interpret and suit the research questions at hand A classificationtree of components is a clear and useful tool in this respect We encourageresearchers to use this approach in further research on tourism expenditure whenbudget share division is fundamental to the researchersrsquo questions

In this study we have encountered two limitations First contrary to studiesthat use macro data in micro data studies the coexistence of full-board halfboard bed and breakfast and accommodation-only tourists makes it impossibleto separate accommodation and food expenditure meaningfully Othersubdivisions (for example discretionary expenditure on activities non-grocery

29Determinants in tourist expenditure composition

shopping and moving around the destination) are technically feasible but canonly increase the percentages of zeros The CODA methodology involves somedegree of amalgamation of components Second even though there are someadvantages to using a database from an official statistics institution as in thiscase (large sample size and scope) the main disadvantage is that the set ofavailable variables cannot be controlled by the researcher A further issue to beaddressed is the time dimension As the EGATUR survey is conducted annuallyfurther research can be done to include a repeat cross-section analysis in orderto capture trends in the effects of predictors

References

Aitchison J (1986) The Statistical Analysis of Compositional Data Monographs on Statistics and AppliedProbability Chapman and Hall London (Reprinted in 2003 with additional material by TheBlackburn Press)

Alegre J Mateo S and Pou LL (2009) lsquoParticipation in tourism consumption and the intensityof participation an analysis of their socio-demographic and economic determinantsrsquo TourismEconomics Vol 15 No 3 pp 531ndash546

Bewley R (1982) lsquoThe generalized addilog demand system applied to Australian time series andcross section datarsquo Australian Economic Papers Vol 21 No 38 pp 177ndash192

Bewley R and Fiebig DG (1988) lsquoEstimation of price elasticities for an international telephonedemand modelrsquo The Journal of Industrial Economics Vol 36 No 4 pp 393ndash409

Cai LA (1998) lsquoAnalysing household food expenditure patterns on trips and vacations a Tobitmodersquo Journal of Hospitality and Tourism Research Vol 22 No 4 pp 338ndash358

Cai LA (1999) lsquoRelationship of household characteristics and lodging expenditure on leisure tripsrsquoJournal of Hospitality and Leisure Marketing Vol 6 No 2 pp 5ndash18

Cai LA Hong G-S and Morrison AM (1995) lsquoHousehold expenditure patterns for tourismproducts and servicesrsquo Journal of Travel and Tourism Marketing Vol 4 No 4 pp 15ndash39

Cannon TF and Ford J (2002) lsquoRelationship of demographic and trip characteristics to visitorspending an analysis of sports travel visitors across timersquo Tourism Economics Vol 8 No 3 pp 263ndash271

Castillo-Manzano JL and Marchena-Goacutemez M (2010) lsquoAnalysis of determinants of airline choiceprofiling the LCC passengerrsquo Applied Economics Letters Vol 18 No 1 pp 49ndash53

Chiou Y-C and Chen Y-H (2010) lsquoFactors influencing the intentions of passengers regarding full ser-vice and low cost carriers a notersquo Journal of Air Transport Management Vol 16 No 4 pp 226ndash228

Coenen M and van Eekeren L (2003) lsquoA study of the demand for domestic tourism by Swedishhouseholds using a two-staged budgeting modelrsquo Scandinavian Journal of Hospitality and TourismVol 3 No 2 pp 114ndash133

Craggs R and Schofield P (2009) lsquoExpenditure-based segmentation and visitor profiling at theQuays in Salford UKrsquo Tourism Economics Vol 15 No 1 pp 243ndash260

Deaton A and Muellbauer J (1980) lsquoAn almost ideal demand systemrsquo American Economic ReviewVol 70 No 3 pp 312ndash326

Dellaert B Borgers A and Timmermans H (1997) lsquoConsumer activity pattern choice devel-opment and test of stage-dependent conjoint choice experimentsrsquo Journal of Retailing and ConsumerServices Vol 4 No 1 pp 25ndash37

Divisekera S (2007) Modelling and Estimation of Tourism Demand Elasticities A Study of TourismExpenditure Allocation in Australia Cooperative Research Centre for Sustainable Tourism (CRCST)Gold Coast

Divisekera S (2009) lsquoEx-post demand for Australian tourism goods and servicesrsquo Tourism EconomicsVol 15 No 1 pp 153ndash180

Divisekera S (2010) lsquoEconomics of leisure and non-leisure tourist demand a study of domesticdemand for Australian tourismrsquo Tourism Economics Vol 16 No 1 pp 117ndash136

Dolnicar S Crouch GI Devinney T Huybers T Louviere JJ and Oppewal H (2008)lsquoTourism and discretionary income allocation Heterogeneity among householdsrsquo TourismManagement Vol 29 No 1 pp 44ndash52

Downward P and Lumsdon L (2000) lsquoThe demand for day-visits an analysis of visitor spendingrsquoTourism Economics Vol 6 No 3 pp 251ndash261

TOURISM ECONOMICS16

The alr is commonly used for statistical modelling and prediction ofcompositions (Fry et al 1996) Conversely the clr transformation is commonlyused for statistical techniques which are based on a metric such as a clusteranalysis because of its preservation of distances even though it leads to asingular covariance matrix Thus while the alr would be appropriate for thepurpose of this article the fact that one component must be used as referencefor all others reduces its flexibility and interpretability Alternatives arepresented at the end of this subsection

While having zero expenditures in absolute data indeed has significantmethodological consequences (Lee 2001) these consequences are arguably moreserious in the CODA methodology If the xid variables contain zeros then log-ratios cannot be computed An obvious initial procedure to reduce zeros is toamalgamate small and conceptually similar components with many zeros intolarger ones In tourism budget research it can be useful to group together allexpenditure on activities or all expenditure on food for instance

In certain instances some zero components result from individual character-istics which are called essential zeros in the CODA literature (Aitchison 1986)Another typology of zeros encountered in the CODA literature is the roundingzero that is a component which is present but is too small to be detected bythe measurement instrument This is typical in chemical biological andgeological compositions and the CODA literature offers ample instruments todeal with rounding zeros

A classic essential zero example in economics is in household budget researchwhen measuring expenditure on tobacco and will essentially be zero if allmembers are non-smokers If tobacco expenditure is decidedly in the researchersrsquointerest the target population should be redefined to include only smokers Inmany instances budget research is often not clear whether zero expenditurescome closer to being essential or rounding zeros In some cases they can beunderstood as corner solutions in a utility maximization problem In others theycan be understood as the inherent randomness of human behaviour or as thelimitations of the data Tourists may spend a certain amount on activities oncertain trips but not on others and so surveys of only one trip will unavoidablycontain some zeros of this type Tourists may also forget or fail to report trivialexpenses such as postcard shopping local bus tickets and going to a museum(see Legoheacuterel (1998) for a discussion on the instability of tourism expenditure)Fry et al (2000) claim that in both situations zeros can be proxied by a verysmall value and thus be treated as rounding zeros In tourist budget researchto treat zero expenses in activities as rounding zeros implies assuming firstthat there is basically no tourist type who will never spend anything onactivities and second that tourists who generally spend little on activities arebasically similar to those who spend nothing or fail to report small expensesWe find both assumptions to be reasonable

Fry et al (2000) essentially used the same zero replacement strategy that waslater suggested by Martiacuten-Fernaacutendez et al (2003) namely replacing xid = 0 with

xprimeid = kδid with 0 lt k lt 1 (4)

where δid is the smallest detectable proportion for individual i and component d

17Determinants in tourist expenditure composition

Martiacuten-Fernaacutendez et al (2003) suggest using k = 065 although a sensitivityanalysis of the results on the choice of k is always advisable (we used k = 030and k = 099 with no sizeable change in the estimates) Next non-zero xid valueshave to be reduced in order to preserve the unit sum and the ratios among non-zero components As suggested by Martiacuten-Fernaacutendez et al (2003) with

xprimeid = xid (1 ndash Σxid=0

xprimeid) (5)

Simulations show this method performs particularly well if the proportion ofzeros is below 10 (Martiacuten-Fernaacutendez et al 2011)

In our data set zeros were present only in one budget category (discretionaryexpenditure) The minimum amount spent by the non-zero group was euro1which roughly corresponds to the price of a city bus ticket the entrance to asubsidized local museum or a cheap souvenir Since the total expenditureis known for each individual we compute δid by dividing euro1 with the totalexpenditure of individual I (see the appendix for the SPSS commandsyntax)

In this article we consider alternative log-ratio transformations which aremore flexible than the alr and clr in that the denominator does not have tobe the same in all ratios This increased flexibility makes it easier to computelog-ratios which are more interpretable with respect to the researchersrsquo questions orhypotheses

In general an interpretable log-ratio transformation is easy to computewhenever there is an interpretable sequential binary partition of componentsinto pairs of groups of components according to the researchersrsquo objectives orto the conceptual similarity of the components These partitions start bydividing components into two clusters and then continue by subdividingone of the clusters into two until each component constitutes its owncluster D components always involve Dndash1 partitions These partitions are bestunderstood as a partition tree or dendrogram (Pawlowsky-Glahn and Egozcue2011)

A meaningful log-ratio transformation takes ratios of the geometric meansof the two component clusters at each partition Numerators and denominatorsare interchangeable In our article we consider xi1 = transportation expenditurexi2 = accommodation and food (basic expenditure) and xi3 = activities andshopping (discretionary expenditure) An interpretable sequential partition isshown in Figure 1

The sequential partition in Figure 1 can have a dual interpretation It mayimply that researchers consider both types of at-destination expenses to bemutually similar and less similar to transportation Or it may imply that theresearch questions involve the distribution of total expenditure betweentransportation and at-destination expenditure and the distribution ofat-destination expenditure into basic and discretionary components as is thecase in our article

The first log-ratio compares transportation expenditure with the geometricmean of accommodation and food (basic expenditure) and activities andshopping (discretionary expenditure) With this ratio we want to observe howtravelling with an LCA or a legacy company affects the share of transportationcompared to non-transportation (at-destination) expenses

TOURISM ECONOMICS18

Figure 1 Sequential partition of total expenditure

⎛ xi1 ⎞ 1 1yi1 = ln ⎜ ndashndashndashndash ⎟ = ln(xi1) ndash ndash ln(xi2) ndash ndash ln(xi3) (6)

⎝ xi2xi3 ⎠ 2 2

Positive values show a transportation share greater than the geometric meanof the remaining two components Negative values show the opposite

The second log-ratio is a ratio of accommodation and food (basic expenditure)over activities and shopping (discretionary expenditure) With this ratio wewant to find out to what tourists allocate the rest of their trip budget andwhether more is allocated to basic or discretionary expenditure once they havepaid for (often in advance) their transportation

⎛ xi2 ⎞yi2 = ln ⎜ ndashndashndash ⎟ = ln(xi2) ndash ln(xi3) (7)

⎝ xi3 ⎠

Positive values show a basic (accommodation and food) share which is largerthan the discretionary (activities and shopping) share Negative values illustratethe opposite

Log-ratio transformations based on sequential binary partitions areproportional to the isometric log-ratio transformation (ilr see Egozcue et al2003 Egozcue and Pawlowsky-Glahn 2005)

As compositions are vector variables they cannot be analysed component-wise by means of univariate regression ANOVA models and the likeSeemingly unrelated regression models for continuous explanatory variables ormultivariate ANOVA (MANOVA) models for categorical explanatory variablesare appropriate MANOVArsquos multivariate tests and statistics (for example

x2 x3 Transportation

Global partition of total expenditure

x1

Accommodation

amp food

(basic)

Activities

amp shopping

(discretionary)

Partition of

expenses other

than transportation

(at-destination

expenses)

19Determinants in tourist expenditure composition

Pillairsquos trace Hotellingrsquos trace or Wilkrsquos lambda) are invariant to howcomponents are arranged in Figure 1 to compute log-ratios (Mateu-Figueras etal 2013) Univariate tests referring to each particular log-ratio are not invari-ant hence the importance of the interpretability of each log-ratio The softwareSPSS 19 is used to estimate the MANOVA model (GLM procedure)

We have to distinguish between three types of variables moderatingexogenous and endogenous to be included in the analysis

Moderating variables are those which modify the effect of exogenousvariables In our article we treat type of airline as a moderator in other wordswe include its interaction terms with all other variables in the MANOVA model

Exogenous variables are assumed to affect expenditure but not the converseSocio-demographic characteristics (travellerrsquos age gender education incomeetc) are obviously determined prior to any travel decision and hence exogenouswith respect to expenditure

Many of the choices tourists make are interdependent (Dellaert et al 1997)or at least planned simultaneously (Fesenmaier and Jeng 2000) So expenditureis arguably decided on at the same time as many trip attributes At least itis unclear whether expenditure is consistently decided on after trip attributesby all travellers We consider them to be endogenous variables and thereforedo not include them in the MANOVA model as explanatory in order to preventendogeneity problems Instead we display them graphically in the log-ratiospace The means of both log-ratios within each endogenous category are thecategory coordinates

The large sample size (see next subsection) makes it possible to use lowp-values Moderating effects with p-values higher than 001 according to eitherPillairsquos trace Hotellingrsquos trace or Wilksrsquo lambda were removed from the modelAll the variables and moderating effects included in the final model aresignificant at 001

Sample and variables

In this article we use secondary official statistics data The data were providedby the Instituto de Estudios Turiacutesticos (IET) an official agency of the Ministryof Industry Energy and Tourism and one which produces the majority oftourism data in Spain The survey is known as the Encuesta de Gasto Turiacutestico(EGATUR) in which tourism expenditure and other tourist information suchas trip information and tourist sociodemographic characteristics are studiedThe EGATUR survey conducted in 27 major Spanish airports in 2010 usedCAPI (computer assisted personal interview) to interview tourists leaving thecountry The sample is non-proportionally stratified by country of residenceairport and month See IET (2012) for further details on the EGATURmethodology

Our universe is a subset of the EGATUR universe which consists ofEuropean leisure visitors arriving by air and spending at least one night inSpain We excluded flights from outside Europe because LCAs mostly operateshort-haul flights We also centred our study on only those trips with one singledestination thus excluding multi-stage trips as the decision process regardingexpenditure composition for these trips is expected to fundamentally differ fromthat of single-stage trips Stays of over 120 days were also excluded

TOURISM ECONOMICS20

For this study we did not consider tourists

bull who have essential zeros in accommodation (tourists who own a house at thedestination or tourists who stay with friends or relatives)

bull who do not decide how much they will spend on certain components(business and study trips)

bull who do not pay for the trip themselves (trips paid for by employers familyfriends prizes offers etc)

bull for whom composition is wholly or partly unobserved (package tourists)

The final sample size was N= 19359From the expenditure variables included in the EGATUR survey database

and used in the model as budget components we first put the amount paidfor transportation (x1) This component has no zeros

Second the amount of money paid for accommodation and food isundistinguishable for full-board half-board or bed and breakfast accommodationTherefore we define a joint accommodation and food component (basic expendi-ture) In this component we include not only the amount paid for consumptionin bars and restaurants but also for buying groceries and everyday products insupermarkets (x2) This component has no zeros

Finally EGATUR provides an aggregated expenditure on activities andshopping (except groceries and everyday products) To this we added theconceptually similar amount paid for moving around the destination (publictransportation andor car rented at the destination) in order to build anactivities and shopping component (x3) This component had 98 zeros whichstresses its discretionary character Zeros were replaced as explained in thestatistical approach subsection

We consider as at-destination (or also called non-transportation) expenses thebasic component (food and accommodation) and the discretionary component(activities and shopping)

Share and log-ratios are described in Table 1 Explanatory variables are thelevel of education income country of residence gender travel groupprofessional status (for pensioners we consider their last professional position)and age The age variable is built as a combination of the original age variableand the variable referring to the individual economic situation This is donein order to have a specific pensioner category taking into account the varyingretirement ages across countries and professions and thus capturing thosetourists who have more free time to undertake a trip regardless of physical ageTable 2 shows their categories and frequencies

As endogenous variables which are not included in the MANOVA modelbut are included in the log-ratio space we include activities undertaken at thedestination accommodation length of stay total expenditure quartilesquartiles of expenditure made at destination per day (basic plus discretionaryexpenses) and a combination of motivation and destination This last variablehas been constructed as a combination of the motivation and the destinationvariable We distinguish between first those tourists travelling for culturaltourism to singular cities second those who come just for leisure either inthe countryside or more commonly at the seaside and third other typologies(Table 3)

21Determinants in tourist expenditure composition

Table 1 Percentage share and log-ratio descriptive statistics

Min Max Mean SD Skewness Kurtosis

Transportation component (x1) 003 9611 2846 1387 0837 0789Basic component (x2) 021 9963 5472 1613 ndash0114 ndash0450Discretionary component (x3) 001 8668 1681 1292 0988 1174Transportationat-destinationlog-ratio (y1) ndash757 548 01506 10702 0714 1404

Basicdiscretionary log-ratio (y2) ndash530 872 17657 18173 1573 2151

Results

Results for the exogenous variables

Table 4 presents the MANOVA results The univariate R2 corrected for thedegrees of freedom is 0045 (first log-ratio) and 0105 (second log-ratio) Themultivariate uncorrected R2 is 1 ndash Λ = 0178 and corrected for the degrees offreedom is 0176 We checked that the results did not change when removingthe four outliers with the highest Cookrsquos distance

Table 4 presents the estimates for the two log-ratios the log-ratio oftransportation over the other two categories (y1) and the log-ratio of basicexpenditure (accommodation and food) over discretionary expenditure (activitiesand shopping) (y2) respectively Because of the moderating effects within eachlog-ratio there are two columns representing the effects when flying with anLCA or with a legacy airline (main and moderating effects have already addedtogether for easier reading) If the LCA and legacy columns are different thereis a significant moderating effect (p-value lt 001)

As we are interpreting log-ratios a positive estimate of a given predictorcategory means that tourists in that category spend more on the numeratorcompared to the denominator than tourists in the reference predictor categoryA negative estimate shows the opposite

The intercept term shows the main effect of company type that is thepredicted log-ratio for each type of airline within the reference category for allvariables (university education medium income resident in the UK or Ireland25ndash44 yearsrsquo old male travelling with partner and mid-level employee) Withinthe reference categories legacy users spend a higher proportion on transportationcompared to other expenses and LCA users spend more on basic expensescompared to discretionary

The level of education seems to have more to do with activities undertakenthan with transportation Results show that a lower level of education resultsin higher expenditure on basic expenses compared to discretionary and is almostequal for users of both types of airline For mainly or only LCA users a lowerlevel of education results in a higher share in transport compared to the othertwo categories If we put it in another way there seems to be a distinct highlyeducated LCA user segment which utilizes the savings in transport toincrease expenditure in non-transportation expenses and even more so indiscretionary

TOURISM ECONOMICS22

Table 2 Frequency distributions of exogenous and moderating variables in the MANOVAmodel

Variables Count Percentage

Level of educationUp to high school 6019 311Universitya 13340 689

Income categoryMediumlow 797 41Mediuma 13186 681Mediumhigh 4376 226High 1000 52

Country of residencePortugal 624 32Other European countries 850 44Belgium 1006 52Netherlands 1069 55Austria Switzerland and Liechtenstein 1122 58France 1326 68Scandinavian countries 1789 92Germany 2363 122Italy 3113 181United Kingdom and Irelanda 6097 315

GenderFemale 8530 441Malea 10829 559

Travel groupAlone 2484 128In family 2980 154With friends 4440 229With partnera 9446 488

Professional statusHomemaker 517 27Unemployed 602 31Student 1432 74Low-level employee 915 47Mid-level employeea 10964 586High-level employee 2138 110Self-employed 2791 144

AgeOver 45 yearsrsquo old and pensioner 1484 77Over 45 yearsrsquo old and not pensioner 5009 25925ndash44 yearsrsquo olda 10569 54615ndash24 yearsrsquo old 2297 119

Type of airline (moderating variable)Legacy 5884 304Low cost 13475 696

Note aReference categories in the MANOVA model chosen either because they are the largestcategories or because they are the most standard conceptually considered

23Determinants in tourist expenditure composition

Table 3 Frequency distributions of endogenous variables (not included in the MANOVAmodel)

Variables Count Percentage

ActivitiesGolfYes 227 12No 19132 988HikingYes 364 19No 18995 981Sporting eventsYes 603 31No 18756 969Nautical sportsYes 1645 85No 17714 915Other sportsYes 904 47No 18455 953Cultural visitsYes 11180 578No 8179 422Cultural eventsYes 2452 127No 16907 873Other cultural activitiesYes 4175 216No 15184 784SpaYes 1199 62No 18160 938Theme parksYes 1745 90No 17614 910GastronomyYes 1565 81No 17794 919Type of accommodationHotel 4ndash5 5056 261Hotel 3 4922 254Hotel lt 3 6673 345Rented appartment 2116 109Other accommodation 592 31Length of stay5 or less nights 9911 5126ndash8 nights 5952 3079ndash12 nights 1592 8213ndash15 nights 1261 6516 and more nights 643 33

TOURISM ECONOMICS24

Table 3 continued

Variables Count Percentage

Total expenditureQuartile 1 5074 262Quartile 2 5104 264Quartile 3 4986 258Quartile 4 4195 217Expenditure at destination per dayQuartile 1 4855 251Quartile 2 4832 250Quartile 3 4868 251Quartile 4 4804 248Motivation-destinationLeisure on coastin countryside 8083 418Urban tourism 5324 275Others 5952 307

Level of income in contrast affects both log-ratios The higher the incomethe lower the transport share The higher the income the lower the basicdiscretionary expense ratio that is a higher portion of the non-transportexpense is devoted to the discretionary budget

As far as the country of residence is concerned the results show that it affectsmainly the basicdiscretionary log-ratio and exhibits strong airline typemoderating effects LCA users tend to differ depending on their country ofresidence when distributing non-transport expenses into basic versusdiscretionary In general when compared to the UK and Ireland residentcategory (reference category) all other European residents have a largerdiscretionary share More precisely the French Italians and Portuguese have thehighest discretionary share within LCA users and other European countries(mainly central and eastern European) have the highest discretionary sharewithin legacy airline users

Gender has a small effect which is constant for both types of airline Femaletravellers tend to slightly increase the transport share and the share of basicexpenses within the non-transport expenses

The travel group seems to affect both log-ratios and has a moderating effectwith airline type Those tourists who travel in a family group have a highershare in transport expenses when flying with legacy airlines In the second log-ratio travelling as a family with legacy airlines increases the basic expensesshare compared to the share in discretionary Travelling with friends increasesthe share in non-transport expenses for both airline types Besides those whotravel with friends spend more in the discretionary share compared to basicexpenses and those who fly with LCA even more so Finally travelling aloneincreases the basic expenses share compared to the share in discretionaryexpenses for both airline types

When professional status is considered the results show that it mostly affectsthe basicdiscretionary log-ratio and there are some relevant airline type

25Determinants in tourist expenditure composition

Table 4 MANOVA results

y1 log-ratio y2 log-ratiotransportation basicat-destination discretionary

LCA Legacy LCA Legacy

Intercept 0225 0274 2195 1881

Up to high school 0192 0078 0406 0437University 0 0 0 0

Mediumlow income 0073 0289 0037 0518Medium income 0 0 0 0Mediumhigh income ndash0227 ndash0058 ndash0269 ndash0006High income ndash0495 ndash0190 ndash0463 ndash0141

Residence in Portugal 0041 0069 ndash1136 ndash0580Residence in other European Countries 0326 ndash0012 ndash0772 ndash0820Residence in Belgium ndash0127 ndash0159 ndash0667 ndash0461Residence in Netherlands ndash0120 ndash0197 ndash0678 ndash0530Residence in Austria + Switzerland +Liechtenstein ndash0017 ndash0087 ndash0687 ndash0436

Residence in France ndash0112 ndash0111 ndash0870 ndash0553Residence in Scandinavian countries ndash0045 ndash0038 ndash0660 ndash0477Residence in Germany ndash0283 ndash0176 ndash0452 ndash0392Residence in Italy ndash0209 ndash0266 ndash0855 ndash0503Residence in UK or Ireland 0 0 0 0

Female 0094 0094 0093 0093Male 0 0 0 0

Travelling alone ndash0053 ndash0002 0081 0126Travelling in family 0085 0193 0014 0213Travelling with friends ndash0237 ndash0191 ndash0504 ndash0282Travelling with partner 0 0 0 0

Homemaker 0080 0160 0253 0235Unemployed 0030 0159 ndash0036 ndash0039Student ndash0005 ndash0104 ndash0331 ndash0624Low-level employee ndash0034 0194 ndash0194 0063Mid-level employee 0 0 0 0High-level employee ndash0008 ndash0064 ndash0001 ndash0139Self-employed 0029 ndash0086 0193 ndash0072

Over 45 yearsrsquo old and pensioner ndash0060 ndash0060 0633 0633Over 45 yearsrsquo old and not pensioner ndash0010 ndash0010 0142 014225ndash44 yearsrsquo old 0 0 0 015ndash24 yearsrsquo old 0002 0002 ndash0019 ndash0019

moderating effects Legacy users with a low professional status (homemakerunemployed or low-level employee) spend more on transport compared to thenon-transport expenses Conversely students using legacy airlines spend moreon non-transport expenses and these non-transport expenses are to a greater

TOURISM ECONOMICS26

Figure 2 Endogenous variables on the log-ratio space Mean log-ratio valueswithin each category of the endogenous variables

extent devoted to discretionary expenses compared to basic expensesHomemakers increase basic expenditure compared to discretionary with respectto the reference category (mid-level employee) for both airline types LCA userswho are self-employed also increase basic expenditure compared to discretionaryLow-level employees flying with an LCA increase the discretionary sharecompared to the basic and high-level employees flying with legacy airlines alsospend relatively more on the discretionary part

The results show that age affects the second log-ratio more than the firstand there is no moderating effect with the airline type Being a pensioner resultsin a higher share in basic expenses compared to discretionary expenses for bothairline types The remaining categories have hardly any differences from thereference 25ndash44 yearsrsquo old

Results for the endogenous variables

As introduced in the methodology section Figure 2 shows how the endogenousvariables behave in the log-ratio space The origin of the graph is representedby the mean log-ratios in Table 1 01506 for the transportother ratio and17657 for the basicdiscretionary ratio and according to this the graph showscategories with higher or lower ratios than the average

075

060

045

030

015

000

ndash015

ndash030

ndash045

ndash060

Ratio tra

nsport

ationa

t-destination e

xpenses

Ratio basicdiscretionary expenses

050 075 100 125 150 175 200 225 250

27Determinants in tourist expenditure composition

Regarding accommodation it seems that rented apartments and other typesof accommodation behave similarly and have the lowest budget share intransport In the case of hotels as the star rating increases expenditure on basiccompared to discretionary increases and expenditure on transport compared toat-destination decreases

As far as the length of stay is concerned longer stays increase the share ofnon-transport expenditure in basic expenses and reduce the share in transport

As regards destination and motivation tourists going to a seaside or country-side destination for leisure purposes behave quite differently from those goingto an urban destination for urban or cultural tourism The former group spendsmore on transport and less on the discretionary portion However those comingfor urban tourism show an increase in the share of non-transport expenses aswell as in the discretionary share compared to basic expenses

The total expenditure variable gives relevant information It is ordered alongthe first log-ratio axis from top to bottom meaning that those tourists havinga lower total expenditure spend relatively more on the transport share and thosewho have a higher total expenditure spend more on the other two parts of thebudget Regarding the expenditure made at destination per day it is ordereddiagonally from the upper-right corner to the bottom-left corner meaning thattourists with a higher at-destination expenditure per day spend less ontransportation and at the same time more on the discretionary part within at-destination expenses

Finally activities undertaken are located in the lower left quadrant of thegraph meaning that undertaking activities (discretionary expenses) decreasesthe share in transport expenses and although they continue to spend more onbasic expenses than on discretionary the discretionary share increases We havehighlighted some of them Those tourists attending sports and cultural eventsare those spending the most on discretionary compared to basic expensesHiking or some nautical sports could be free of charge thus tourists doing theseactivities have a basic to discretionary ratio close to the mean As for thetransportation share all activities tend to reduce this share about equally withthe exception of golf which leads to a much more substantial reduction

Conclusion and discussion

The main purpose of this article was to study the composition of touristexpenditure and its determinants that is the drivers of the share of touristexpenditure allocated to the different categories of a travel budget and withspecial emphasis on the distinction between legacy versus LCA travellers

The main differences between legacy and LCA users are that income somecountries of residence the travel group and some occupations have differenteffects depending on the type of airline Tourists with a medium or low incomewho travel with legacy airlines tend to spend a greater share of their travelbudget on transportation and basic expenses but those travelling with LCAstend to spend more at the destination and more specifically ondiscretionary expenditure LCA tourists residing in the lsquoother Europeancountriesrsquo spend more on transportation Families flying with legacy airlinesspend more on the transportation share compared to the at-destination share

TOURISM ECONOMICS28

and also spend more on basic expenses compared to discretionary ones Finallyin terms of occupational status legacy users with low-level employment spendmore on transportation whereas the opposite occurs with LCA travellers whospend more on the discretionary component Students and high-level employeesflying with legacy airlines spend more on the discretionary part of the tripbudget

As far as implications for management are concerned assuming that theobjectives of DMOs are to increase the allocation to the at-destinationcomponent of the trip budget (that is lowering the first log-ratio) some of therecommended actions would be to increase marketing efforts directed atmedium- and high-income earners especially at those travelling on LCAs toincrease marketing strategies in some of the European outbound markets suchas Germany (especially those using LCA) and Italy and the Netherlands(especially those flying with legacy airlines) DMOs should also focus on thosetourists travelling with friends (airline type is inconsequential) and studentsusing legacy companies As for the endogenous variables DMOs should directtheir efforts towards those tourists undertaking activities towards those stayinga little bit longer than a week and towards those with high total expenditureFurthermore if DMOs are also interested in capturing the groups who spendrelatively more on discretionary expenses (lowering the second log-ratio) theyshould focus their marketing efforts on those tourists who are highly educatedas well as those who have medium to high incomes and who use LCAsAdditionally they should focus on markets other than the UK and IrelandYoung tourists and tourists who travel with friends (where airline type isinconsequential) low-level employees and students using LCAs as well asstudents and high-level employees flying on legacy airlines should also betargeted In relation to the endogenous variables DMOs must take into accountthose tourists undertaking payable activities those on shorter stays or comingfor urban and cultural tourism or other purposes

The appeal of the CODA methodology for studying tourism budgets lies inthe fact that once the variables have been transformed the researcher can usestandard and well understood statistical models with unbounded error termdistributions while ensuring that the results will be compositionally coherentand that the standard statistical assumptions will hold Conversely results arenot compositionally coherent when we use the MANOVA model on raw sharewithout transforming it into log-ratios We have seen with our data that 84of individuals included in the sample have at least one limit of the 90 shareprediction intervals outside the [01] range

Contrary to the classic and highly restrictive alr and clr transformations theCODA methodology offers the potential to construct tailor-made log-ratios whichare intuitive to interpret and suit the research questions at hand A classificationtree of components is a clear and useful tool in this respect We encourageresearchers to use this approach in further research on tourism expenditure whenbudget share division is fundamental to the researchersrsquo questions

In this study we have encountered two limitations First contrary to studiesthat use macro data in micro data studies the coexistence of full-board halfboard bed and breakfast and accommodation-only tourists makes it impossibleto separate accommodation and food expenditure meaningfully Othersubdivisions (for example discretionary expenditure on activities non-grocery

29Determinants in tourist expenditure composition

shopping and moving around the destination) are technically feasible but canonly increase the percentages of zeros The CODA methodology involves somedegree of amalgamation of components Second even though there are someadvantages to using a database from an official statistics institution as in thiscase (large sample size and scope) the main disadvantage is that the set ofavailable variables cannot be controlled by the researcher A further issue to beaddressed is the time dimension As the EGATUR survey is conducted annuallyfurther research can be done to include a repeat cross-section analysis in orderto capture trends in the effects of predictors

References

Aitchison J (1986) The Statistical Analysis of Compositional Data Monographs on Statistics and AppliedProbability Chapman and Hall London (Reprinted in 2003 with additional material by TheBlackburn Press)

Alegre J Mateo S and Pou LL (2009) lsquoParticipation in tourism consumption and the intensityof participation an analysis of their socio-demographic and economic determinantsrsquo TourismEconomics Vol 15 No 3 pp 531ndash546

Bewley R (1982) lsquoThe generalized addilog demand system applied to Australian time series andcross section datarsquo Australian Economic Papers Vol 21 No 38 pp 177ndash192

Bewley R and Fiebig DG (1988) lsquoEstimation of price elasticities for an international telephonedemand modelrsquo The Journal of Industrial Economics Vol 36 No 4 pp 393ndash409

Cai LA (1998) lsquoAnalysing household food expenditure patterns on trips and vacations a Tobitmodersquo Journal of Hospitality and Tourism Research Vol 22 No 4 pp 338ndash358

Cai LA (1999) lsquoRelationship of household characteristics and lodging expenditure on leisure tripsrsquoJournal of Hospitality and Leisure Marketing Vol 6 No 2 pp 5ndash18

Cai LA Hong G-S and Morrison AM (1995) lsquoHousehold expenditure patterns for tourismproducts and servicesrsquo Journal of Travel and Tourism Marketing Vol 4 No 4 pp 15ndash39

Cannon TF and Ford J (2002) lsquoRelationship of demographic and trip characteristics to visitorspending an analysis of sports travel visitors across timersquo Tourism Economics Vol 8 No 3 pp 263ndash271

Castillo-Manzano JL and Marchena-Goacutemez M (2010) lsquoAnalysis of determinants of airline choiceprofiling the LCC passengerrsquo Applied Economics Letters Vol 18 No 1 pp 49ndash53

Chiou Y-C and Chen Y-H (2010) lsquoFactors influencing the intentions of passengers regarding full ser-vice and low cost carriers a notersquo Journal of Air Transport Management Vol 16 No 4 pp 226ndash228

Coenen M and van Eekeren L (2003) lsquoA study of the demand for domestic tourism by Swedishhouseholds using a two-staged budgeting modelrsquo Scandinavian Journal of Hospitality and TourismVol 3 No 2 pp 114ndash133

Craggs R and Schofield P (2009) lsquoExpenditure-based segmentation and visitor profiling at theQuays in Salford UKrsquo Tourism Economics Vol 15 No 1 pp 243ndash260

Deaton A and Muellbauer J (1980) lsquoAn almost ideal demand systemrsquo American Economic ReviewVol 70 No 3 pp 312ndash326

Dellaert B Borgers A and Timmermans H (1997) lsquoConsumer activity pattern choice devel-opment and test of stage-dependent conjoint choice experimentsrsquo Journal of Retailing and ConsumerServices Vol 4 No 1 pp 25ndash37

Divisekera S (2007) Modelling and Estimation of Tourism Demand Elasticities A Study of TourismExpenditure Allocation in Australia Cooperative Research Centre for Sustainable Tourism (CRCST)Gold Coast

Divisekera S (2009) lsquoEx-post demand for Australian tourism goods and servicesrsquo Tourism EconomicsVol 15 No 1 pp 153ndash180

Divisekera S (2010) lsquoEconomics of leisure and non-leisure tourist demand a study of domesticdemand for Australian tourismrsquo Tourism Economics Vol 16 No 1 pp 117ndash136

Dolnicar S Crouch GI Devinney T Huybers T Louviere JJ and Oppewal H (2008)lsquoTourism and discretionary income allocation Heterogeneity among householdsrsquo TourismManagement Vol 29 No 1 pp 44ndash52

Downward P and Lumsdon L (2000) lsquoThe demand for day-visits an analysis of visitor spendingrsquoTourism Economics Vol 6 No 3 pp 251ndash261

17Determinants in tourist expenditure composition

Martiacuten-Fernaacutendez et al (2003) suggest using k = 065 although a sensitivityanalysis of the results on the choice of k is always advisable (we used k = 030and k = 099 with no sizeable change in the estimates) Next non-zero xid valueshave to be reduced in order to preserve the unit sum and the ratios among non-zero components As suggested by Martiacuten-Fernaacutendez et al (2003) with

xprimeid = xid (1 ndash Σxid=0

xprimeid) (5)

Simulations show this method performs particularly well if the proportion ofzeros is below 10 (Martiacuten-Fernaacutendez et al 2011)

In our data set zeros were present only in one budget category (discretionaryexpenditure) The minimum amount spent by the non-zero group was euro1which roughly corresponds to the price of a city bus ticket the entrance to asubsidized local museum or a cheap souvenir Since the total expenditureis known for each individual we compute δid by dividing euro1 with the totalexpenditure of individual I (see the appendix for the SPSS commandsyntax)

In this article we consider alternative log-ratio transformations which aremore flexible than the alr and clr in that the denominator does not have tobe the same in all ratios This increased flexibility makes it easier to computelog-ratios which are more interpretable with respect to the researchersrsquo questions orhypotheses

In general an interpretable log-ratio transformation is easy to computewhenever there is an interpretable sequential binary partition of componentsinto pairs of groups of components according to the researchersrsquo objectives orto the conceptual similarity of the components These partitions start bydividing components into two clusters and then continue by subdividingone of the clusters into two until each component constitutes its owncluster D components always involve Dndash1 partitions These partitions are bestunderstood as a partition tree or dendrogram (Pawlowsky-Glahn and Egozcue2011)

A meaningful log-ratio transformation takes ratios of the geometric meansof the two component clusters at each partition Numerators and denominatorsare interchangeable In our article we consider xi1 = transportation expenditurexi2 = accommodation and food (basic expenditure) and xi3 = activities andshopping (discretionary expenditure) An interpretable sequential partition isshown in Figure 1

The sequential partition in Figure 1 can have a dual interpretation It mayimply that researchers consider both types of at-destination expenses to bemutually similar and less similar to transportation Or it may imply that theresearch questions involve the distribution of total expenditure betweentransportation and at-destination expenditure and the distribution ofat-destination expenditure into basic and discretionary components as is thecase in our article

The first log-ratio compares transportation expenditure with the geometricmean of accommodation and food (basic expenditure) and activities andshopping (discretionary expenditure) With this ratio we want to observe howtravelling with an LCA or a legacy company affects the share of transportationcompared to non-transportation (at-destination) expenses

TOURISM ECONOMICS18

Figure 1 Sequential partition of total expenditure

⎛ xi1 ⎞ 1 1yi1 = ln ⎜ ndashndashndashndash ⎟ = ln(xi1) ndash ndash ln(xi2) ndash ndash ln(xi3) (6)

⎝ xi2xi3 ⎠ 2 2

Positive values show a transportation share greater than the geometric meanof the remaining two components Negative values show the opposite

The second log-ratio is a ratio of accommodation and food (basic expenditure)over activities and shopping (discretionary expenditure) With this ratio wewant to find out to what tourists allocate the rest of their trip budget andwhether more is allocated to basic or discretionary expenditure once they havepaid for (often in advance) their transportation

⎛ xi2 ⎞yi2 = ln ⎜ ndashndashndash ⎟ = ln(xi2) ndash ln(xi3) (7)

⎝ xi3 ⎠

Positive values show a basic (accommodation and food) share which is largerthan the discretionary (activities and shopping) share Negative values illustratethe opposite

Log-ratio transformations based on sequential binary partitions areproportional to the isometric log-ratio transformation (ilr see Egozcue et al2003 Egozcue and Pawlowsky-Glahn 2005)

As compositions are vector variables they cannot be analysed component-wise by means of univariate regression ANOVA models and the likeSeemingly unrelated regression models for continuous explanatory variables ormultivariate ANOVA (MANOVA) models for categorical explanatory variablesare appropriate MANOVArsquos multivariate tests and statistics (for example

x2 x3 Transportation

Global partition of total expenditure

x1

Accommodation

amp food

(basic)

Activities

amp shopping

(discretionary)

Partition of

expenses other

than transportation

(at-destination

expenses)

19Determinants in tourist expenditure composition

Pillairsquos trace Hotellingrsquos trace or Wilkrsquos lambda) are invariant to howcomponents are arranged in Figure 1 to compute log-ratios (Mateu-Figueras etal 2013) Univariate tests referring to each particular log-ratio are not invari-ant hence the importance of the interpretability of each log-ratio The softwareSPSS 19 is used to estimate the MANOVA model (GLM procedure)

We have to distinguish between three types of variables moderatingexogenous and endogenous to be included in the analysis

Moderating variables are those which modify the effect of exogenousvariables In our article we treat type of airline as a moderator in other wordswe include its interaction terms with all other variables in the MANOVA model

Exogenous variables are assumed to affect expenditure but not the converseSocio-demographic characteristics (travellerrsquos age gender education incomeetc) are obviously determined prior to any travel decision and hence exogenouswith respect to expenditure

Many of the choices tourists make are interdependent (Dellaert et al 1997)or at least planned simultaneously (Fesenmaier and Jeng 2000) So expenditureis arguably decided on at the same time as many trip attributes At least itis unclear whether expenditure is consistently decided on after trip attributesby all travellers We consider them to be endogenous variables and thereforedo not include them in the MANOVA model as explanatory in order to preventendogeneity problems Instead we display them graphically in the log-ratiospace The means of both log-ratios within each endogenous category are thecategory coordinates

The large sample size (see next subsection) makes it possible to use lowp-values Moderating effects with p-values higher than 001 according to eitherPillairsquos trace Hotellingrsquos trace or Wilksrsquo lambda were removed from the modelAll the variables and moderating effects included in the final model aresignificant at 001

Sample and variables

In this article we use secondary official statistics data The data were providedby the Instituto de Estudios Turiacutesticos (IET) an official agency of the Ministryof Industry Energy and Tourism and one which produces the majority oftourism data in Spain The survey is known as the Encuesta de Gasto Turiacutestico(EGATUR) in which tourism expenditure and other tourist information suchas trip information and tourist sociodemographic characteristics are studiedThe EGATUR survey conducted in 27 major Spanish airports in 2010 usedCAPI (computer assisted personal interview) to interview tourists leaving thecountry The sample is non-proportionally stratified by country of residenceairport and month See IET (2012) for further details on the EGATURmethodology

Our universe is a subset of the EGATUR universe which consists ofEuropean leisure visitors arriving by air and spending at least one night inSpain We excluded flights from outside Europe because LCAs mostly operateshort-haul flights We also centred our study on only those trips with one singledestination thus excluding multi-stage trips as the decision process regardingexpenditure composition for these trips is expected to fundamentally differ fromthat of single-stage trips Stays of over 120 days were also excluded

TOURISM ECONOMICS20

For this study we did not consider tourists

bull who have essential zeros in accommodation (tourists who own a house at thedestination or tourists who stay with friends or relatives)

bull who do not decide how much they will spend on certain components(business and study trips)

bull who do not pay for the trip themselves (trips paid for by employers familyfriends prizes offers etc)

bull for whom composition is wholly or partly unobserved (package tourists)

The final sample size was N= 19359From the expenditure variables included in the EGATUR survey database

and used in the model as budget components we first put the amount paidfor transportation (x1) This component has no zeros

Second the amount of money paid for accommodation and food isundistinguishable for full-board half-board or bed and breakfast accommodationTherefore we define a joint accommodation and food component (basic expendi-ture) In this component we include not only the amount paid for consumptionin bars and restaurants but also for buying groceries and everyday products insupermarkets (x2) This component has no zeros

Finally EGATUR provides an aggregated expenditure on activities andshopping (except groceries and everyday products) To this we added theconceptually similar amount paid for moving around the destination (publictransportation andor car rented at the destination) in order to build anactivities and shopping component (x3) This component had 98 zeros whichstresses its discretionary character Zeros were replaced as explained in thestatistical approach subsection

We consider as at-destination (or also called non-transportation) expenses thebasic component (food and accommodation) and the discretionary component(activities and shopping)

Share and log-ratios are described in Table 1 Explanatory variables are thelevel of education income country of residence gender travel groupprofessional status (for pensioners we consider their last professional position)and age The age variable is built as a combination of the original age variableand the variable referring to the individual economic situation This is donein order to have a specific pensioner category taking into account the varyingretirement ages across countries and professions and thus capturing thosetourists who have more free time to undertake a trip regardless of physical ageTable 2 shows their categories and frequencies

As endogenous variables which are not included in the MANOVA modelbut are included in the log-ratio space we include activities undertaken at thedestination accommodation length of stay total expenditure quartilesquartiles of expenditure made at destination per day (basic plus discretionaryexpenses) and a combination of motivation and destination This last variablehas been constructed as a combination of the motivation and the destinationvariable We distinguish between first those tourists travelling for culturaltourism to singular cities second those who come just for leisure either inthe countryside or more commonly at the seaside and third other typologies(Table 3)

21Determinants in tourist expenditure composition

Table 1 Percentage share and log-ratio descriptive statistics

Min Max Mean SD Skewness Kurtosis

Transportation component (x1) 003 9611 2846 1387 0837 0789Basic component (x2) 021 9963 5472 1613 ndash0114 ndash0450Discretionary component (x3) 001 8668 1681 1292 0988 1174Transportationat-destinationlog-ratio (y1) ndash757 548 01506 10702 0714 1404

Basicdiscretionary log-ratio (y2) ndash530 872 17657 18173 1573 2151

Results

Results for the exogenous variables

Table 4 presents the MANOVA results The univariate R2 corrected for thedegrees of freedom is 0045 (first log-ratio) and 0105 (second log-ratio) Themultivariate uncorrected R2 is 1 ndash Λ = 0178 and corrected for the degrees offreedom is 0176 We checked that the results did not change when removingthe four outliers with the highest Cookrsquos distance

Table 4 presents the estimates for the two log-ratios the log-ratio oftransportation over the other two categories (y1) and the log-ratio of basicexpenditure (accommodation and food) over discretionary expenditure (activitiesand shopping) (y2) respectively Because of the moderating effects within eachlog-ratio there are two columns representing the effects when flying with anLCA or with a legacy airline (main and moderating effects have already addedtogether for easier reading) If the LCA and legacy columns are different thereis a significant moderating effect (p-value lt 001)

As we are interpreting log-ratios a positive estimate of a given predictorcategory means that tourists in that category spend more on the numeratorcompared to the denominator than tourists in the reference predictor categoryA negative estimate shows the opposite

The intercept term shows the main effect of company type that is thepredicted log-ratio for each type of airline within the reference category for allvariables (university education medium income resident in the UK or Ireland25ndash44 yearsrsquo old male travelling with partner and mid-level employee) Withinthe reference categories legacy users spend a higher proportion on transportationcompared to other expenses and LCA users spend more on basic expensescompared to discretionary

The level of education seems to have more to do with activities undertakenthan with transportation Results show that a lower level of education resultsin higher expenditure on basic expenses compared to discretionary and is almostequal for users of both types of airline For mainly or only LCA users a lowerlevel of education results in a higher share in transport compared to the othertwo categories If we put it in another way there seems to be a distinct highlyeducated LCA user segment which utilizes the savings in transport toincrease expenditure in non-transportation expenses and even more so indiscretionary

TOURISM ECONOMICS22

Table 2 Frequency distributions of exogenous and moderating variables in the MANOVAmodel

Variables Count Percentage

Level of educationUp to high school 6019 311Universitya 13340 689

Income categoryMediumlow 797 41Mediuma 13186 681Mediumhigh 4376 226High 1000 52

Country of residencePortugal 624 32Other European countries 850 44Belgium 1006 52Netherlands 1069 55Austria Switzerland and Liechtenstein 1122 58France 1326 68Scandinavian countries 1789 92Germany 2363 122Italy 3113 181United Kingdom and Irelanda 6097 315

GenderFemale 8530 441Malea 10829 559

Travel groupAlone 2484 128In family 2980 154With friends 4440 229With partnera 9446 488

Professional statusHomemaker 517 27Unemployed 602 31Student 1432 74Low-level employee 915 47Mid-level employeea 10964 586High-level employee 2138 110Self-employed 2791 144

AgeOver 45 yearsrsquo old and pensioner 1484 77Over 45 yearsrsquo old and not pensioner 5009 25925ndash44 yearsrsquo olda 10569 54615ndash24 yearsrsquo old 2297 119

Type of airline (moderating variable)Legacy 5884 304Low cost 13475 696

Note aReference categories in the MANOVA model chosen either because they are the largestcategories or because they are the most standard conceptually considered

23Determinants in tourist expenditure composition

Table 3 Frequency distributions of endogenous variables (not included in the MANOVAmodel)

Variables Count Percentage

ActivitiesGolfYes 227 12No 19132 988HikingYes 364 19No 18995 981Sporting eventsYes 603 31No 18756 969Nautical sportsYes 1645 85No 17714 915Other sportsYes 904 47No 18455 953Cultural visitsYes 11180 578No 8179 422Cultural eventsYes 2452 127No 16907 873Other cultural activitiesYes 4175 216No 15184 784SpaYes 1199 62No 18160 938Theme parksYes 1745 90No 17614 910GastronomyYes 1565 81No 17794 919Type of accommodationHotel 4ndash5 5056 261Hotel 3 4922 254Hotel lt 3 6673 345Rented appartment 2116 109Other accommodation 592 31Length of stay5 or less nights 9911 5126ndash8 nights 5952 3079ndash12 nights 1592 8213ndash15 nights 1261 6516 and more nights 643 33

TOURISM ECONOMICS24

Table 3 continued

Variables Count Percentage

Total expenditureQuartile 1 5074 262Quartile 2 5104 264Quartile 3 4986 258Quartile 4 4195 217Expenditure at destination per dayQuartile 1 4855 251Quartile 2 4832 250Quartile 3 4868 251Quartile 4 4804 248Motivation-destinationLeisure on coastin countryside 8083 418Urban tourism 5324 275Others 5952 307

Level of income in contrast affects both log-ratios The higher the incomethe lower the transport share The higher the income the lower the basicdiscretionary expense ratio that is a higher portion of the non-transportexpense is devoted to the discretionary budget

As far as the country of residence is concerned the results show that it affectsmainly the basicdiscretionary log-ratio and exhibits strong airline typemoderating effects LCA users tend to differ depending on their country ofresidence when distributing non-transport expenses into basic versusdiscretionary In general when compared to the UK and Ireland residentcategory (reference category) all other European residents have a largerdiscretionary share More precisely the French Italians and Portuguese have thehighest discretionary share within LCA users and other European countries(mainly central and eastern European) have the highest discretionary sharewithin legacy airline users

Gender has a small effect which is constant for both types of airline Femaletravellers tend to slightly increase the transport share and the share of basicexpenses within the non-transport expenses

The travel group seems to affect both log-ratios and has a moderating effectwith airline type Those tourists who travel in a family group have a highershare in transport expenses when flying with legacy airlines In the second log-ratio travelling as a family with legacy airlines increases the basic expensesshare compared to the share in discretionary Travelling with friends increasesthe share in non-transport expenses for both airline types Besides those whotravel with friends spend more in the discretionary share compared to basicexpenses and those who fly with LCA even more so Finally travelling aloneincreases the basic expenses share compared to the share in discretionaryexpenses for both airline types

When professional status is considered the results show that it mostly affectsthe basicdiscretionary log-ratio and there are some relevant airline type

25Determinants in tourist expenditure composition

Table 4 MANOVA results

y1 log-ratio y2 log-ratiotransportation basicat-destination discretionary

LCA Legacy LCA Legacy

Intercept 0225 0274 2195 1881

Up to high school 0192 0078 0406 0437University 0 0 0 0

Mediumlow income 0073 0289 0037 0518Medium income 0 0 0 0Mediumhigh income ndash0227 ndash0058 ndash0269 ndash0006High income ndash0495 ndash0190 ndash0463 ndash0141

Residence in Portugal 0041 0069 ndash1136 ndash0580Residence in other European Countries 0326 ndash0012 ndash0772 ndash0820Residence in Belgium ndash0127 ndash0159 ndash0667 ndash0461Residence in Netherlands ndash0120 ndash0197 ndash0678 ndash0530Residence in Austria + Switzerland +Liechtenstein ndash0017 ndash0087 ndash0687 ndash0436

Residence in France ndash0112 ndash0111 ndash0870 ndash0553Residence in Scandinavian countries ndash0045 ndash0038 ndash0660 ndash0477Residence in Germany ndash0283 ndash0176 ndash0452 ndash0392Residence in Italy ndash0209 ndash0266 ndash0855 ndash0503Residence in UK or Ireland 0 0 0 0

Female 0094 0094 0093 0093Male 0 0 0 0

Travelling alone ndash0053 ndash0002 0081 0126Travelling in family 0085 0193 0014 0213Travelling with friends ndash0237 ndash0191 ndash0504 ndash0282Travelling with partner 0 0 0 0

Homemaker 0080 0160 0253 0235Unemployed 0030 0159 ndash0036 ndash0039Student ndash0005 ndash0104 ndash0331 ndash0624Low-level employee ndash0034 0194 ndash0194 0063Mid-level employee 0 0 0 0High-level employee ndash0008 ndash0064 ndash0001 ndash0139Self-employed 0029 ndash0086 0193 ndash0072

Over 45 yearsrsquo old and pensioner ndash0060 ndash0060 0633 0633Over 45 yearsrsquo old and not pensioner ndash0010 ndash0010 0142 014225ndash44 yearsrsquo old 0 0 0 015ndash24 yearsrsquo old 0002 0002 ndash0019 ndash0019

moderating effects Legacy users with a low professional status (homemakerunemployed or low-level employee) spend more on transport compared to thenon-transport expenses Conversely students using legacy airlines spend moreon non-transport expenses and these non-transport expenses are to a greater

TOURISM ECONOMICS26

Figure 2 Endogenous variables on the log-ratio space Mean log-ratio valueswithin each category of the endogenous variables

extent devoted to discretionary expenses compared to basic expensesHomemakers increase basic expenditure compared to discretionary with respectto the reference category (mid-level employee) for both airline types LCA userswho are self-employed also increase basic expenditure compared to discretionaryLow-level employees flying with an LCA increase the discretionary sharecompared to the basic and high-level employees flying with legacy airlines alsospend relatively more on the discretionary part

The results show that age affects the second log-ratio more than the firstand there is no moderating effect with the airline type Being a pensioner resultsin a higher share in basic expenses compared to discretionary expenses for bothairline types The remaining categories have hardly any differences from thereference 25ndash44 yearsrsquo old

Results for the endogenous variables

As introduced in the methodology section Figure 2 shows how the endogenousvariables behave in the log-ratio space The origin of the graph is representedby the mean log-ratios in Table 1 01506 for the transportother ratio and17657 for the basicdiscretionary ratio and according to this the graph showscategories with higher or lower ratios than the average

075

060

045

030

015

000

ndash015

ndash030

ndash045

ndash060

Ratio tra

nsport

ationa

t-destination e

xpenses

Ratio basicdiscretionary expenses

050 075 100 125 150 175 200 225 250

27Determinants in tourist expenditure composition

Regarding accommodation it seems that rented apartments and other typesof accommodation behave similarly and have the lowest budget share intransport In the case of hotels as the star rating increases expenditure on basiccompared to discretionary increases and expenditure on transport compared toat-destination decreases

As far as the length of stay is concerned longer stays increase the share ofnon-transport expenditure in basic expenses and reduce the share in transport

As regards destination and motivation tourists going to a seaside or country-side destination for leisure purposes behave quite differently from those goingto an urban destination for urban or cultural tourism The former group spendsmore on transport and less on the discretionary portion However those comingfor urban tourism show an increase in the share of non-transport expenses aswell as in the discretionary share compared to basic expenses

The total expenditure variable gives relevant information It is ordered alongthe first log-ratio axis from top to bottom meaning that those tourists havinga lower total expenditure spend relatively more on the transport share and thosewho have a higher total expenditure spend more on the other two parts of thebudget Regarding the expenditure made at destination per day it is ordereddiagonally from the upper-right corner to the bottom-left corner meaning thattourists with a higher at-destination expenditure per day spend less ontransportation and at the same time more on the discretionary part within at-destination expenses

Finally activities undertaken are located in the lower left quadrant of thegraph meaning that undertaking activities (discretionary expenses) decreasesthe share in transport expenses and although they continue to spend more onbasic expenses than on discretionary the discretionary share increases We havehighlighted some of them Those tourists attending sports and cultural eventsare those spending the most on discretionary compared to basic expensesHiking or some nautical sports could be free of charge thus tourists doing theseactivities have a basic to discretionary ratio close to the mean As for thetransportation share all activities tend to reduce this share about equally withthe exception of golf which leads to a much more substantial reduction

Conclusion and discussion

The main purpose of this article was to study the composition of touristexpenditure and its determinants that is the drivers of the share of touristexpenditure allocated to the different categories of a travel budget and withspecial emphasis on the distinction between legacy versus LCA travellers

The main differences between legacy and LCA users are that income somecountries of residence the travel group and some occupations have differenteffects depending on the type of airline Tourists with a medium or low incomewho travel with legacy airlines tend to spend a greater share of their travelbudget on transportation and basic expenses but those travelling with LCAstend to spend more at the destination and more specifically ondiscretionary expenditure LCA tourists residing in the lsquoother Europeancountriesrsquo spend more on transportation Families flying with legacy airlinesspend more on the transportation share compared to the at-destination share

TOURISM ECONOMICS28

and also spend more on basic expenses compared to discretionary ones Finallyin terms of occupational status legacy users with low-level employment spendmore on transportation whereas the opposite occurs with LCA travellers whospend more on the discretionary component Students and high-level employeesflying with legacy airlines spend more on the discretionary part of the tripbudget

As far as implications for management are concerned assuming that theobjectives of DMOs are to increase the allocation to the at-destinationcomponent of the trip budget (that is lowering the first log-ratio) some of therecommended actions would be to increase marketing efforts directed atmedium- and high-income earners especially at those travelling on LCAs toincrease marketing strategies in some of the European outbound markets suchas Germany (especially those using LCA) and Italy and the Netherlands(especially those flying with legacy airlines) DMOs should also focus on thosetourists travelling with friends (airline type is inconsequential) and studentsusing legacy companies As for the endogenous variables DMOs should directtheir efforts towards those tourists undertaking activities towards those stayinga little bit longer than a week and towards those with high total expenditureFurthermore if DMOs are also interested in capturing the groups who spendrelatively more on discretionary expenses (lowering the second log-ratio) theyshould focus their marketing efforts on those tourists who are highly educatedas well as those who have medium to high incomes and who use LCAsAdditionally they should focus on markets other than the UK and IrelandYoung tourists and tourists who travel with friends (where airline type isinconsequential) low-level employees and students using LCAs as well asstudents and high-level employees flying on legacy airlines should also betargeted In relation to the endogenous variables DMOs must take into accountthose tourists undertaking payable activities those on shorter stays or comingfor urban and cultural tourism or other purposes

The appeal of the CODA methodology for studying tourism budgets lies inthe fact that once the variables have been transformed the researcher can usestandard and well understood statistical models with unbounded error termdistributions while ensuring that the results will be compositionally coherentand that the standard statistical assumptions will hold Conversely results arenot compositionally coherent when we use the MANOVA model on raw sharewithout transforming it into log-ratios We have seen with our data that 84of individuals included in the sample have at least one limit of the 90 shareprediction intervals outside the [01] range

Contrary to the classic and highly restrictive alr and clr transformations theCODA methodology offers the potential to construct tailor-made log-ratios whichare intuitive to interpret and suit the research questions at hand A classificationtree of components is a clear and useful tool in this respect We encourageresearchers to use this approach in further research on tourism expenditure whenbudget share division is fundamental to the researchersrsquo questions

In this study we have encountered two limitations First contrary to studiesthat use macro data in micro data studies the coexistence of full-board halfboard bed and breakfast and accommodation-only tourists makes it impossibleto separate accommodation and food expenditure meaningfully Othersubdivisions (for example discretionary expenditure on activities non-grocery

29Determinants in tourist expenditure composition

shopping and moving around the destination) are technically feasible but canonly increase the percentages of zeros The CODA methodology involves somedegree of amalgamation of components Second even though there are someadvantages to using a database from an official statistics institution as in thiscase (large sample size and scope) the main disadvantage is that the set ofavailable variables cannot be controlled by the researcher A further issue to beaddressed is the time dimension As the EGATUR survey is conducted annuallyfurther research can be done to include a repeat cross-section analysis in orderto capture trends in the effects of predictors

References

Aitchison J (1986) The Statistical Analysis of Compositional Data Monographs on Statistics and AppliedProbability Chapman and Hall London (Reprinted in 2003 with additional material by TheBlackburn Press)

Alegre J Mateo S and Pou LL (2009) lsquoParticipation in tourism consumption and the intensityof participation an analysis of their socio-demographic and economic determinantsrsquo TourismEconomics Vol 15 No 3 pp 531ndash546

Bewley R (1982) lsquoThe generalized addilog demand system applied to Australian time series andcross section datarsquo Australian Economic Papers Vol 21 No 38 pp 177ndash192

Bewley R and Fiebig DG (1988) lsquoEstimation of price elasticities for an international telephonedemand modelrsquo The Journal of Industrial Economics Vol 36 No 4 pp 393ndash409

Cai LA (1998) lsquoAnalysing household food expenditure patterns on trips and vacations a Tobitmodersquo Journal of Hospitality and Tourism Research Vol 22 No 4 pp 338ndash358

Cai LA (1999) lsquoRelationship of household characteristics and lodging expenditure on leisure tripsrsquoJournal of Hospitality and Leisure Marketing Vol 6 No 2 pp 5ndash18

Cai LA Hong G-S and Morrison AM (1995) lsquoHousehold expenditure patterns for tourismproducts and servicesrsquo Journal of Travel and Tourism Marketing Vol 4 No 4 pp 15ndash39

Cannon TF and Ford J (2002) lsquoRelationship of demographic and trip characteristics to visitorspending an analysis of sports travel visitors across timersquo Tourism Economics Vol 8 No 3 pp 263ndash271

Castillo-Manzano JL and Marchena-Goacutemez M (2010) lsquoAnalysis of determinants of airline choiceprofiling the LCC passengerrsquo Applied Economics Letters Vol 18 No 1 pp 49ndash53

Chiou Y-C and Chen Y-H (2010) lsquoFactors influencing the intentions of passengers regarding full ser-vice and low cost carriers a notersquo Journal of Air Transport Management Vol 16 No 4 pp 226ndash228

Coenen M and van Eekeren L (2003) lsquoA study of the demand for domestic tourism by Swedishhouseholds using a two-staged budgeting modelrsquo Scandinavian Journal of Hospitality and TourismVol 3 No 2 pp 114ndash133

Craggs R and Schofield P (2009) lsquoExpenditure-based segmentation and visitor profiling at theQuays in Salford UKrsquo Tourism Economics Vol 15 No 1 pp 243ndash260

Deaton A and Muellbauer J (1980) lsquoAn almost ideal demand systemrsquo American Economic ReviewVol 70 No 3 pp 312ndash326

Dellaert B Borgers A and Timmermans H (1997) lsquoConsumer activity pattern choice devel-opment and test of stage-dependent conjoint choice experimentsrsquo Journal of Retailing and ConsumerServices Vol 4 No 1 pp 25ndash37

Divisekera S (2007) Modelling and Estimation of Tourism Demand Elasticities A Study of TourismExpenditure Allocation in Australia Cooperative Research Centre for Sustainable Tourism (CRCST)Gold Coast

Divisekera S (2009) lsquoEx-post demand for Australian tourism goods and servicesrsquo Tourism EconomicsVol 15 No 1 pp 153ndash180

Divisekera S (2010) lsquoEconomics of leisure and non-leisure tourist demand a study of domesticdemand for Australian tourismrsquo Tourism Economics Vol 16 No 1 pp 117ndash136

Dolnicar S Crouch GI Devinney T Huybers T Louviere JJ and Oppewal H (2008)lsquoTourism and discretionary income allocation Heterogeneity among householdsrsquo TourismManagement Vol 29 No 1 pp 44ndash52

Downward P and Lumsdon L (2000) lsquoThe demand for day-visits an analysis of visitor spendingrsquoTourism Economics Vol 6 No 3 pp 251ndash261

TOURISM ECONOMICS18

Figure 1 Sequential partition of total expenditure

⎛ xi1 ⎞ 1 1yi1 = ln ⎜ ndashndashndashndash ⎟ = ln(xi1) ndash ndash ln(xi2) ndash ndash ln(xi3) (6)

⎝ xi2xi3 ⎠ 2 2

Positive values show a transportation share greater than the geometric meanof the remaining two components Negative values show the opposite

The second log-ratio is a ratio of accommodation and food (basic expenditure)over activities and shopping (discretionary expenditure) With this ratio wewant to find out to what tourists allocate the rest of their trip budget andwhether more is allocated to basic or discretionary expenditure once they havepaid for (often in advance) their transportation

⎛ xi2 ⎞yi2 = ln ⎜ ndashndashndash ⎟ = ln(xi2) ndash ln(xi3) (7)

⎝ xi3 ⎠

Positive values show a basic (accommodation and food) share which is largerthan the discretionary (activities and shopping) share Negative values illustratethe opposite

Log-ratio transformations based on sequential binary partitions areproportional to the isometric log-ratio transformation (ilr see Egozcue et al2003 Egozcue and Pawlowsky-Glahn 2005)

As compositions are vector variables they cannot be analysed component-wise by means of univariate regression ANOVA models and the likeSeemingly unrelated regression models for continuous explanatory variables ormultivariate ANOVA (MANOVA) models for categorical explanatory variablesare appropriate MANOVArsquos multivariate tests and statistics (for example

x2 x3 Transportation

Global partition of total expenditure

x1

Accommodation

amp food

(basic)

Activities

amp shopping

(discretionary)

Partition of

expenses other

than transportation

(at-destination

expenses)

19Determinants in tourist expenditure composition

Pillairsquos trace Hotellingrsquos trace or Wilkrsquos lambda) are invariant to howcomponents are arranged in Figure 1 to compute log-ratios (Mateu-Figueras etal 2013) Univariate tests referring to each particular log-ratio are not invari-ant hence the importance of the interpretability of each log-ratio The softwareSPSS 19 is used to estimate the MANOVA model (GLM procedure)

We have to distinguish between three types of variables moderatingexogenous and endogenous to be included in the analysis

Moderating variables are those which modify the effect of exogenousvariables In our article we treat type of airline as a moderator in other wordswe include its interaction terms with all other variables in the MANOVA model

Exogenous variables are assumed to affect expenditure but not the converseSocio-demographic characteristics (travellerrsquos age gender education incomeetc) are obviously determined prior to any travel decision and hence exogenouswith respect to expenditure

Many of the choices tourists make are interdependent (Dellaert et al 1997)or at least planned simultaneously (Fesenmaier and Jeng 2000) So expenditureis arguably decided on at the same time as many trip attributes At least itis unclear whether expenditure is consistently decided on after trip attributesby all travellers We consider them to be endogenous variables and thereforedo not include them in the MANOVA model as explanatory in order to preventendogeneity problems Instead we display them graphically in the log-ratiospace The means of both log-ratios within each endogenous category are thecategory coordinates

The large sample size (see next subsection) makes it possible to use lowp-values Moderating effects with p-values higher than 001 according to eitherPillairsquos trace Hotellingrsquos trace or Wilksrsquo lambda were removed from the modelAll the variables and moderating effects included in the final model aresignificant at 001

Sample and variables

In this article we use secondary official statistics data The data were providedby the Instituto de Estudios Turiacutesticos (IET) an official agency of the Ministryof Industry Energy and Tourism and one which produces the majority oftourism data in Spain The survey is known as the Encuesta de Gasto Turiacutestico(EGATUR) in which tourism expenditure and other tourist information suchas trip information and tourist sociodemographic characteristics are studiedThe EGATUR survey conducted in 27 major Spanish airports in 2010 usedCAPI (computer assisted personal interview) to interview tourists leaving thecountry The sample is non-proportionally stratified by country of residenceairport and month See IET (2012) for further details on the EGATURmethodology

Our universe is a subset of the EGATUR universe which consists ofEuropean leisure visitors arriving by air and spending at least one night inSpain We excluded flights from outside Europe because LCAs mostly operateshort-haul flights We also centred our study on only those trips with one singledestination thus excluding multi-stage trips as the decision process regardingexpenditure composition for these trips is expected to fundamentally differ fromthat of single-stage trips Stays of over 120 days were also excluded

TOURISM ECONOMICS20

For this study we did not consider tourists

bull who have essential zeros in accommodation (tourists who own a house at thedestination or tourists who stay with friends or relatives)

bull who do not decide how much they will spend on certain components(business and study trips)

bull who do not pay for the trip themselves (trips paid for by employers familyfriends prizes offers etc)

bull for whom composition is wholly or partly unobserved (package tourists)

The final sample size was N= 19359From the expenditure variables included in the EGATUR survey database

and used in the model as budget components we first put the amount paidfor transportation (x1) This component has no zeros

Second the amount of money paid for accommodation and food isundistinguishable for full-board half-board or bed and breakfast accommodationTherefore we define a joint accommodation and food component (basic expendi-ture) In this component we include not only the amount paid for consumptionin bars and restaurants but also for buying groceries and everyday products insupermarkets (x2) This component has no zeros

Finally EGATUR provides an aggregated expenditure on activities andshopping (except groceries and everyday products) To this we added theconceptually similar amount paid for moving around the destination (publictransportation andor car rented at the destination) in order to build anactivities and shopping component (x3) This component had 98 zeros whichstresses its discretionary character Zeros were replaced as explained in thestatistical approach subsection

We consider as at-destination (or also called non-transportation) expenses thebasic component (food and accommodation) and the discretionary component(activities and shopping)

Share and log-ratios are described in Table 1 Explanatory variables are thelevel of education income country of residence gender travel groupprofessional status (for pensioners we consider their last professional position)and age The age variable is built as a combination of the original age variableand the variable referring to the individual economic situation This is donein order to have a specific pensioner category taking into account the varyingretirement ages across countries and professions and thus capturing thosetourists who have more free time to undertake a trip regardless of physical ageTable 2 shows their categories and frequencies

As endogenous variables which are not included in the MANOVA modelbut are included in the log-ratio space we include activities undertaken at thedestination accommodation length of stay total expenditure quartilesquartiles of expenditure made at destination per day (basic plus discretionaryexpenses) and a combination of motivation and destination This last variablehas been constructed as a combination of the motivation and the destinationvariable We distinguish between first those tourists travelling for culturaltourism to singular cities second those who come just for leisure either inthe countryside or more commonly at the seaside and third other typologies(Table 3)

21Determinants in tourist expenditure composition

Table 1 Percentage share and log-ratio descriptive statistics

Min Max Mean SD Skewness Kurtosis

Transportation component (x1) 003 9611 2846 1387 0837 0789Basic component (x2) 021 9963 5472 1613 ndash0114 ndash0450Discretionary component (x3) 001 8668 1681 1292 0988 1174Transportationat-destinationlog-ratio (y1) ndash757 548 01506 10702 0714 1404

Basicdiscretionary log-ratio (y2) ndash530 872 17657 18173 1573 2151

Results

Results for the exogenous variables

Table 4 presents the MANOVA results The univariate R2 corrected for thedegrees of freedom is 0045 (first log-ratio) and 0105 (second log-ratio) Themultivariate uncorrected R2 is 1 ndash Λ = 0178 and corrected for the degrees offreedom is 0176 We checked that the results did not change when removingthe four outliers with the highest Cookrsquos distance

Table 4 presents the estimates for the two log-ratios the log-ratio oftransportation over the other two categories (y1) and the log-ratio of basicexpenditure (accommodation and food) over discretionary expenditure (activitiesand shopping) (y2) respectively Because of the moderating effects within eachlog-ratio there are two columns representing the effects when flying with anLCA or with a legacy airline (main and moderating effects have already addedtogether for easier reading) If the LCA and legacy columns are different thereis a significant moderating effect (p-value lt 001)

As we are interpreting log-ratios a positive estimate of a given predictorcategory means that tourists in that category spend more on the numeratorcompared to the denominator than tourists in the reference predictor categoryA negative estimate shows the opposite

The intercept term shows the main effect of company type that is thepredicted log-ratio for each type of airline within the reference category for allvariables (university education medium income resident in the UK or Ireland25ndash44 yearsrsquo old male travelling with partner and mid-level employee) Withinthe reference categories legacy users spend a higher proportion on transportationcompared to other expenses and LCA users spend more on basic expensescompared to discretionary

The level of education seems to have more to do with activities undertakenthan with transportation Results show that a lower level of education resultsin higher expenditure on basic expenses compared to discretionary and is almostequal for users of both types of airline For mainly or only LCA users a lowerlevel of education results in a higher share in transport compared to the othertwo categories If we put it in another way there seems to be a distinct highlyeducated LCA user segment which utilizes the savings in transport toincrease expenditure in non-transportation expenses and even more so indiscretionary

TOURISM ECONOMICS22

Table 2 Frequency distributions of exogenous and moderating variables in the MANOVAmodel

Variables Count Percentage

Level of educationUp to high school 6019 311Universitya 13340 689

Income categoryMediumlow 797 41Mediuma 13186 681Mediumhigh 4376 226High 1000 52

Country of residencePortugal 624 32Other European countries 850 44Belgium 1006 52Netherlands 1069 55Austria Switzerland and Liechtenstein 1122 58France 1326 68Scandinavian countries 1789 92Germany 2363 122Italy 3113 181United Kingdom and Irelanda 6097 315

GenderFemale 8530 441Malea 10829 559

Travel groupAlone 2484 128In family 2980 154With friends 4440 229With partnera 9446 488

Professional statusHomemaker 517 27Unemployed 602 31Student 1432 74Low-level employee 915 47Mid-level employeea 10964 586High-level employee 2138 110Self-employed 2791 144

AgeOver 45 yearsrsquo old and pensioner 1484 77Over 45 yearsrsquo old and not pensioner 5009 25925ndash44 yearsrsquo olda 10569 54615ndash24 yearsrsquo old 2297 119

Type of airline (moderating variable)Legacy 5884 304Low cost 13475 696

Note aReference categories in the MANOVA model chosen either because they are the largestcategories or because they are the most standard conceptually considered

23Determinants in tourist expenditure composition

Table 3 Frequency distributions of endogenous variables (not included in the MANOVAmodel)

Variables Count Percentage

ActivitiesGolfYes 227 12No 19132 988HikingYes 364 19No 18995 981Sporting eventsYes 603 31No 18756 969Nautical sportsYes 1645 85No 17714 915Other sportsYes 904 47No 18455 953Cultural visitsYes 11180 578No 8179 422Cultural eventsYes 2452 127No 16907 873Other cultural activitiesYes 4175 216No 15184 784SpaYes 1199 62No 18160 938Theme parksYes 1745 90No 17614 910GastronomyYes 1565 81No 17794 919Type of accommodationHotel 4ndash5 5056 261Hotel 3 4922 254Hotel lt 3 6673 345Rented appartment 2116 109Other accommodation 592 31Length of stay5 or less nights 9911 5126ndash8 nights 5952 3079ndash12 nights 1592 8213ndash15 nights 1261 6516 and more nights 643 33

TOURISM ECONOMICS24

Table 3 continued

Variables Count Percentage

Total expenditureQuartile 1 5074 262Quartile 2 5104 264Quartile 3 4986 258Quartile 4 4195 217Expenditure at destination per dayQuartile 1 4855 251Quartile 2 4832 250Quartile 3 4868 251Quartile 4 4804 248Motivation-destinationLeisure on coastin countryside 8083 418Urban tourism 5324 275Others 5952 307

Level of income in contrast affects both log-ratios The higher the incomethe lower the transport share The higher the income the lower the basicdiscretionary expense ratio that is a higher portion of the non-transportexpense is devoted to the discretionary budget

As far as the country of residence is concerned the results show that it affectsmainly the basicdiscretionary log-ratio and exhibits strong airline typemoderating effects LCA users tend to differ depending on their country ofresidence when distributing non-transport expenses into basic versusdiscretionary In general when compared to the UK and Ireland residentcategory (reference category) all other European residents have a largerdiscretionary share More precisely the French Italians and Portuguese have thehighest discretionary share within LCA users and other European countries(mainly central and eastern European) have the highest discretionary sharewithin legacy airline users

Gender has a small effect which is constant for both types of airline Femaletravellers tend to slightly increase the transport share and the share of basicexpenses within the non-transport expenses

The travel group seems to affect both log-ratios and has a moderating effectwith airline type Those tourists who travel in a family group have a highershare in transport expenses when flying with legacy airlines In the second log-ratio travelling as a family with legacy airlines increases the basic expensesshare compared to the share in discretionary Travelling with friends increasesthe share in non-transport expenses for both airline types Besides those whotravel with friends spend more in the discretionary share compared to basicexpenses and those who fly with LCA even more so Finally travelling aloneincreases the basic expenses share compared to the share in discretionaryexpenses for both airline types

When professional status is considered the results show that it mostly affectsthe basicdiscretionary log-ratio and there are some relevant airline type

25Determinants in tourist expenditure composition

Table 4 MANOVA results

y1 log-ratio y2 log-ratiotransportation basicat-destination discretionary

LCA Legacy LCA Legacy

Intercept 0225 0274 2195 1881

Up to high school 0192 0078 0406 0437University 0 0 0 0

Mediumlow income 0073 0289 0037 0518Medium income 0 0 0 0Mediumhigh income ndash0227 ndash0058 ndash0269 ndash0006High income ndash0495 ndash0190 ndash0463 ndash0141

Residence in Portugal 0041 0069 ndash1136 ndash0580Residence in other European Countries 0326 ndash0012 ndash0772 ndash0820Residence in Belgium ndash0127 ndash0159 ndash0667 ndash0461Residence in Netherlands ndash0120 ndash0197 ndash0678 ndash0530Residence in Austria + Switzerland +Liechtenstein ndash0017 ndash0087 ndash0687 ndash0436

Residence in France ndash0112 ndash0111 ndash0870 ndash0553Residence in Scandinavian countries ndash0045 ndash0038 ndash0660 ndash0477Residence in Germany ndash0283 ndash0176 ndash0452 ndash0392Residence in Italy ndash0209 ndash0266 ndash0855 ndash0503Residence in UK or Ireland 0 0 0 0

Female 0094 0094 0093 0093Male 0 0 0 0

Travelling alone ndash0053 ndash0002 0081 0126Travelling in family 0085 0193 0014 0213Travelling with friends ndash0237 ndash0191 ndash0504 ndash0282Travelling with partner 0 0 0 0

Homemaker 0080 0160 0253 0235Unemployed 0030 0159 ndash0036 ndash0039Student ndash0005 ndash0104 ndash0331 ndash0624Low-level employee ndash0034 0194 ndash0194 0063Mid-level employee 0 0 0 0High-level employee ndash0008 ndash0064 ndash0001 ndash0139Self-employed 0029 ndash0086 0193 ndash0072

Over 45 yearsrsquo old and pensioner ndash0060 ndash0060 0633 0633Over 45 yearsrsquo old and not pensioner ndash0010 ndash0010 0142 014225ndash44 yearsrsquo old 0 0 0 015ndash24 yearsrsquo old 0002 0002 ndash0019 ndash0019

moderating effects Legacy users with a low professional status (homemakerunemployed or low-level employee) spend more on transport compared to thenon-transport expenses Conversely students using legacy airlines spend moreon non-transport expenses and these non-transport expenses are to a greater

TOURISM ECONOMICS26

Figure 2 Endogenous variables on the log-ratio space Mean log-ratio valueswithin each category of the endogenous variables

extent devoted to discretionary expenses compared to basic expensesHomemakers increase basic expenditure compared to discretionary with respectto the reference category (mid-level employee) for both airline types LCA userswho are self-employed also increase basic expenditure compared to discretionaryLow-level employees flying with an LCA increase the discretionary sharecompared to the basic and high-level employees flying with legacy airlines alsospend relatively more on the discretionary part

The results show that age affects the second log-ratio more than the firstand there is no moderating effect with the airline type Being a pensioner resultsin a higher share in basic expenses compared to discretionary expenses for bothairline types The remaining categories have hardly any differences from thereference 25ndash44 yearsrsquo old

Results for the endogenous variables

As introduced in the methodology section Figure 2 shows how the endogenousvariables behave in the log-ratio space The origin of the graph is representedby the mean log-ratios in Table 1 01506 for the transportother ratio and17657 for the basicdiscretionary ratio and according to this the graph showscategories with higher or lower ratios than the average

075

060

045

030

015

000

ndash015

ndash030

ndash045

ndash060

Ratio tra

nsport

ationa

t-destination e

xpenses

Ratio basicdiscretionary expenses

050 075 100 125 150 175 200 225 250

27Determinants in tourist expenditure composition

Regarding accommodation it seems that rented apartments and other typesof accommodation behave similarly and have the lowest budget share intransport In the case of hotels as the star rating increases expenditure on basiccompared to discretionary increases and expenditure on transport compared toat-destination decreases

As far as the length of stay is concerned longer stays increase the share ofnon-transport expenditure in basic expenses and reduce the share in transport

As regards destination and motivation tourists going to a seaside or country-side destination for leisure purposes behave quite differently from those goingto an urban destination for urban or cultural tourism The former group spendsmore on transport and less on the discretionary portion However those comingfor urban tourism show an increase in the share of non-transport expenses aswell as in the discretionary share compared to basic expenses

The total expenditure variable gives relevant information It is ordered alongthe first log-ratio axis from top to bottom meaning that those tourists havinga lower total expenditure spend relatively more on the transport share and thosewho have a higher total expenditure spend more on the other two parts of thebudget Regarding the expenditure made at destination per day it is ordereddiagonally from the upper-right corner to the bottom-left corner meaning thattourists with a higher at-destination expenditure per day spend less ontransportation and at the same time more on the discretionary part within at-destination expenses

Finally activities undertaken are located in the lower left quadrant of thegraph meaning that undertaking activities (discretionary expenses) decreasesthe share in transport expenses and although they continue to spend more onbasic expenses than on discretionary the discretionary share increases We havehighlighted some of them Those tourists attending sports and cultural eventsare those spending the most on discretionary compared to basic expensesHiking or some nautical sports could be free of charge thus tourists doing theseactivities have a basic to discretionary ratio close to the mean As for thetransportation share all activities tend to reduce this share about equally withthe exception of golf which leads to a much more substantial reduction

Conclusion and discussion

The main purpose of this article was to study the composition of touristexpenditure and its determinants that is the drivers of the share of touristexpenditure allocated to the different categories of a travel budget and withspecial emphasis on the distinction between legacy versus LCA travellers

The main differences between legacy and LCA users are that income somecountries of residence the travel group and some occupations have differenteffects depending on the type of airline Tourists with a medium or low incomewho travel with legacy airlines tend to spend a greater share of their travelbudget on transportation and basic expenses but those travelling with LCAstend to spend more at the destination and more specifically ondiscretionary expenditure LCA tourists residing in the lsquoother Europeancountriesrsquo spend more on transportation Families flying with legacy airlinesspend more on the transportation share compared to the at-destination share

TOURISM ECONOMICS28

and also spend more on basic expenses compared to discretionary ones Finallyin terms of occupational status legacy users with low-level employment spendmore on transportation whereas the opposite occurs with LCA travellers whospend more on the discretionary component Students and high-level employeesflying with legacy airlines spend more on the discretionary part of the tripbudget

As far as implications for management are concerned assuming that theobjectives of DMOs are to increase the allocation to the at-destinationcomponent of the trip budget (that is lowering the first log-ratio) some of therecommended actions would be to increase marketing efforts directed atmedium- and high-income earners especially at those travelling on LCAs toincrease marketing strategies in some of the European outbound markets suchas Germany (especially those using LCA) and Italy and the Netherlands(especially those flying with legacy airlines) DMOs should also focus on thosetourists travelling with friends (airline type is inconsequential) and studentsusing legacy companies As for the endogenous variables DMOs should directtheir efforts towards those tourists undertaking activities towards those stayinga little bit longer than a week and towards those with high total expenditureFurthermore if DMOs are also interested in capturing the groups who spendrelatively more on discretionary expenses (lowering the second log-ratio) theyshould focus their marketing efforts on those tourists who are highly educatedas well as those who have medium to high incomes and who use LCAsAdditionally they should focus on markets other than the UK and IrelandYoung tourists and tourists who travel with friends (where airline type isinconsequential) low-level employees and students using LCAs as well asstudents and high-level employees flying on legacy airlines should also betargeted In relation to the endogenous variables DMOs must take into accountthose tourists undertaking payable activities those on shorter stays or comingfor urban and cultural tourism or other purposes

The appeal of the CODA methodology for studying tourism budgets lies inthe fact that once the variables have been transformed the researcher can usestandard and well understood statistical models with unbounded error termdistributions while ensuring that the results will be compositionally coherentand that the standard statistical assumptions will hold Conversely results arenot compositionally coherent when we use the MANOVA model on raw sharewithout transforming it into log-ratios We have seen with our data that 84of individuals included in the sample have at least one limit of the 90 shareprediction intervals outside the [01] range

Contrary to the classic and highly restrictive alr and clr transformations theCODA methodology offers the potential to construct tailor-made log-ratios whichare intuitive to interpret and suit the research questions at hand A classificationtree of components is a clear and useful tool in this respect We encourageresearchers to use this approach in further research on tourism expenditure whenbudget share division is fundamental to the researchersrsquo questions

In this study we have encountered two limitations First contrary to studiesthat use macro data in micro data studies the coexistence of full-board halfboard bed and breakfast and accommodation-only tourists makes it impossibleto separate accommodation and food expenditure meaningfully Othersubdivisions (for example discretionary expenditure on activities non-grocery

29Determinants in tourist expenditure composition

shopping and moving around the destination) are technically feasible but canonly increase the percentages of zeros The CODA methodology involves somedegree of amalgamation of components Second even though there are someadvantages to using a database from an official statistics institution as in thiscase (large sample size and scope) the main disadvantage is that the set ofavailable variables cannot be controlled by the researcher A further issue to beaddressed is the time dimension As the EGATUR survey is conducted annuallyfurther research can be done to include a repeat cross-section analysis in orderto capture trends in the effects of predictors

References

Aitchison J (1986) The Statistical Analysis of Compositional Data Monographs on Statistics and AppliedProbability Chapman and Hall London (Reprinted in 2003 with additional material by TheBlackburn Press)

Alegre J Mateo S and Pou LL (2009) lsquoParticipation in tourism consumption and the intensityof participation an analysis of their socio-demographic and economic determinantsrsquo TourismEconomics Vol 15 No 3 pp 531ndash546

Bewley R (1982) lsquoThe generalized addilog demand system applied to Australian time series andcross section datarsquo Australian Economic Papers Vol 21 No 38 pp 177ndash192

Bewley R and Fiebig DG (1988) lsquoEstimation of price elasticities for an international telephonedemand modelrsquo The Journal of Industrial Economics Vol 36 No 4 pp 393ndash409

Cai LA (1998) lsquoAnalysing household food expenditure patterns on trips and vacations a Tobitmodersquo Journal of Hospitality and Tourism Research Vol 22 No 4 pp 338ndash358

Cai LA (1999) lsquoRelationship of household characteristics and lodging expenditure on leisure tripsrsquoJournal of Hospitality and Leisure Marketing Vol 6 No 2 pp 5ndash18

Cai LA Hong G-S and Morrison AM (1995) lsquoHousehold expenditure patterns for tourismproducts and servicesrsquo Journal of Travel and Tourism Marketing Vol 4 No 4 pp 15ndash39

Cannon TF and Ford J (2002) lsquoRelationship of demographic and trip characteristics to visitorspending an analysis of sports travel visitors across timersquo Tourism Economics Vol 8 No 3 pp 263ndash271

Castillo-Manzano JL and Marchena-Goacutemez M (2010) lsquoAnalysis of determinants of airline choiceprofiling the LCC passengerrsquo Applied Economics Letters Vol 18 No 1 pp 49ndash53

Chiou Y-C and Chen Y-H (2010) lsquoFactors influencing the intentions of passengers regarding full ser-vice and low cost carriers a notersquo Journal of Air Transport Management Vol 16 No 4 pp 226ndash228

Coenen M and van Eekeren L (2003) lsquoA study of the demand for domestic tourism by Swedishhouseholds using a two-staged budgeting modelrsquo Scandinavian Journal of Hospitality and TourismVol 3 No 2 pp 114ndash133

Craggs R and Schofield P (2009) lsquoExpenditure-based segmentation and visitor profiling at theQuays in Salford UKrsquo Tourism Economics Vol 15 No 1 pp 243ndash260

Deaton A and Muellbauer J (1980) lsquoAn almost ideal demand systemrsquo American Economic ReviewVol 70 No 3 pp 312ndash326

Dellaert B Borgers A and Timmermans H (1997) lsquoConsumer activity pattern choice devel-opment and test of stage-dependent conjoint choice experimentsrsquo Journal of Retailing and ConsumerServices Vol 4 No 1 pp 25ndash37

Divisekera S (2007) Modelling and Estimation of Tourism Demand Elasticities A Study of TourismExpenditure Allocation in Australia Cooperative Research Centre for Sustainable Tourism (CRCST)Gold Coast

Divisekera S (2009) lsquoEx-post demand for Australian tourism goods and servicesrsquo Tourism EconomicsVol 15 No 1 pp 153ndash180

Divisekera S (2010) lsquoEconomics of leisure and non-leisure tourist demand a study of domesticdemand for Australian tourismrsquo Tourism Economics Vol 16 No 1 pp 117ndash136

Dolnicar S Crouch GI Devinney T Huybers T Louviere JJ and Oppewal H (2008)lsquoTourism and discretionary income allocation Heterogeneity among householdsrsquo TourismManagement Vol 29 No 1 pp 44ndash52

Downward P and Lumsdon L (2000) lsquoThe demand for day-visits an analysis of visitor spendingrsquoTourism Economics Vol 6 No 3 pp 251ndash261

19Determinants in tourist expenditure composition

Pillairsquos trace Hotellingrsquos trace or Wilkrsquos lambda) are invariant to howcomponents are arranged in Figure 1 to compute log-ratios (Mateu-Figueras etal 2013) Univariate tests referring to each particular log-ratio are not invari-ant hence the importance of the interpretability of each log-ratio The softwareSPSS 19 is used to estimate the MANOVA model (GLM procedure)

We have to distinguish between three types of variables moderatingexogenous and endogenous to be included in the analysis

Moderating variables are those which modify the effect of exogenousvariables In our article we treat type of airline as a moderator in other wordswe include its interaction terms with all other variables in the MANOVA model

Exogenous variables are assumed to affect expenditure but not the converseSocio-demographic characteristics (travellerrsquos age gender education incomeetc) are obviously determined prior to any travel decision and hence exogenouswith respect to expenditure

Many of the choices tourists make are interdependent (Dellaert et al 1997)or at least planned simultaneously (Fesenmaier and Jeng 2000) So expenditureis arguably decided on at the same time as many trip attributes At least itis unclear whether expenditure is consistently decided on after trip attributesby all travellers We consider them to be endogenous variables and thereforedo not include them in the MANOVA model as explanatory in order to preventendogeneity problems Instead we display them graphically in the log-ratiospace The means of both log-ratios within each endogenous category are thecategory coordinates

The large sample size (see next subsection) makes it possible to use lowp-values Moderating effects with p-values higher than 001 according to eitherPillairsquos trace Hotellingrsquos trace or Wilksrsquo lambda were removed from the modelAll the variables and moderating effects included in the final model aresignificant at 001

Sample and variables

In this article we use secondary official statistics data The data were providedby the Instituto de Estudios Turiacutesticos (IET) an official agency of the Ministryof Industry Energy and Tourism and one which produces the majority oftourism data in Spain The survey is known as the Encuesta de Gasto Turiacutestico(EGATUR) in which tourism expenditure and other tourist information suchas trip information and tourist sociodemographic characteristics are studiedThe EGATUR survey conducted in 27 major Spanish airports in 2010 usedCAPI (computer assisted personal interview) to interview tourists leaving thecountry The sample is non-proportionally stratified by country of residenceairport and month See IET (2012) for further details on the EGATURmethodology

Our universe is a subset of the EGATUR universe which consists ofEuropean leisure visitors arriving by air and spending at least one night inSpain We excluded flights from outside Europe because LCAs mostly operateshort-haul flights We also centred our study on only those trips with one singledestination thus excluding multi-stage trips as the decision process regardingexpenditure composition for these trips is expected to fundamentally differ fromthat of single-stage trips Stays of over 120 days were also excluded

TOURISM ECONOMICS20

For this study we did not consider tourists

bull who have essential zeros in accommodation (tourists who own a house at thedestination or tourists who stay with friends or relatives)

bull who do not decide how much they will spend on certain components(business and study trips)

bull who do not pay for the trip themselves (trips paid for by employers familyfriends prizes offers etc)

bull for whom composition is wholly or partly unobserved (package tourists)

The final sample size was N= 19359From the expenditure variables included in the EGATUR survey database

and used in the model as budget components we first put the amount paidfor transportation (x1) This component has no zeros

Second the amount of money paid for accommodation and food isundistinguishable for full-board half-board or bed and breakfast accommodationTherefore we define a joint accommodation and food component (basic expendi-ture) In this component we include not only the amount paid for consumptionin bars and restaurants but also for buying groceries and everyday products insupermarkets (x2) This component has no zeros

Finally EGATUR provides an aggregated expenditure on activities andshopping (except groceries and everyday products) To this we added theconceptually similar amount paid for moving around the destination (publictransportation andor car rented at the destination) in order to build anactivities and shopping component (x3) This component had 98 zeros whichstresses its discretionary character Zeros were replaced as explained in thestatistical approach subsection

We consider as at-destination (or also called non-transportation) expenses thebasic component (food and accommodation) and the discretionary component(activities and shopping)

Share and log-ratios are described in Table 1 Explanatory variables are thelevel of education income country of residence gender travel groupprofessional status (for pensioners we consider their last professional position)and age The age variable is built as a combination of the original age variableand the variable referring to the individual economic situation This is donein order to have a specific pensioner category taking into account the varyingretirement ages across countries and professions and thus capturing thosetourists who have more free time to undertake a trip regardless of physical ageTable 2 shows their categories and frequencies

As endogenous variables which are not included in the MANOVA modelbut are included in the log-ratio space we include activities undertaken at thedestination accommodation length of stay total expenditure quartilesquartiles of expenditure made at destination per day (basic plus discretionaryexpenses) and a combination of motivation and destination This last variablehas been constructed as a combination of the motivation and the destinationvariable We distinguish between first those tourists travelling for culturaltourism to singular cities second those who come just for leisure either inthe countryside or more commonly at the seaside and third other typologies(Table 3)

21Determinants in tourist expenditure composition

Table 1 Percentage share and log-ratio descriptive statistics

Min Max Mean SD Skewness Kurtosis

Transportation component (x1) 003 9611 2846 1387 0837 0789Basic component (x2) 021 9963 5472 1613 ndash0114 ndash0450Discretionary component (x3) 001 8668 1681 1292 0988 1174Transportationat-destinationlog-ratio (y1) ndash757 548 01506 10702 0714 1404

Basicdiscretionary log-ratio (y2) ndash530 872 17657 18173 1573 2151

Results

Results for the exogenous variables

Table 4 presents the MANOVA results The univariate R2 corrected for thedegrees of freedom is 0045 (first log-ratio) and 0105 (second log-ratio) Themultivariate uncorrected R2 is 1 ndash Λ = 0178 and corrected for the degrees offreedom is 0176 We checked that the results did not change when removingthe four outliers with the highest Cookrsquos distance

Table 4 presents the estimates for the two log-ratios the log-ratio oftransportation over the other two categories (y1) and the log-ratio of basicexpenditure (accommodation and food) over discretionary expenditure (activitiesand shopping) (y2) respectively Because of the moderating effects within eachlog-ratio there are two columns representing the effects when flying with anLCA or with a legacy airline (main and moderating effects have already addedtogether for easier reading) If the LCA and legacy columns are different thereis a significant moderating effect (p-value lt 001)

As we are interpreting log-ratios a positive estimate of a given predictorcategory means that tourists in that category spend more on the numeratorcompared to the denominator than tourists in the reference predictor categoryA negative estimate shows the opposite

The intercept term shows the main effect of company type that is thepredicted log-ratio for each type of airline within the reference category for allvariables (university education medium income resident in the UK or Ireland25ndash44 yearsrsquo old male travelling with partner and mid-level employee) Withinthe reference categories legacy users spend a higher proportion on transportationcompared to other expenses and LCA users spend more on basic expensescompared to discretionary

The level of education seems to have more to do with activities undertakenthan with transportation Results show that a lower level of education resultsin higher expenditure on basic expenses compared to discretionary and is almostequal for users of both types of airline For mainly or only LCA users a lowerlevel of education results in a higher share in transport compared to the othertwo categories If we put it in another way there seems to be a distinct highlyeducated LCA user segment which utilizes the savings in transport toincrease expenditure in non-transportation expenses and even more so indiscretionary

TOURISM ECONOMICS22

Table 2 Frequency distributions of exogenous and moderating variables in the MANOVAmodel

Variables Count Percentage

Level of educationUp to high school 6019 311Universitya 13340 689

Income categoryMediumlow 797 41Mediuma 13186 681Mediumhigh 4376 226High 1000 52

Country of residencePortugal 624 32Other European countries 850 44Belgium 1006 52Netherlands 1069 55Austria Switzerland and Liechtenstein 1122 58France 1326 68Scandinavian countries 1789 92Germany 2363 122Italy 3113 181United Kingdom and Irelanda 6097 315

GenderFemale 8530 441Malea 10829 559

Travel groupAlone 2484 128In family 2980 154With friends 4440 229With partnera 9446 488

Professional statusHomemaker 517 27Unemployed 602 31Student 1432 74Low-level employee 915 47Mid-level employeea 10964 586High-level employee 2138 110Self-employed 2791 144

AgeOver 45 yearsrsquo old and pensioner 1484 77Over 45 yearsrsquo old and not pensioner 5009 25925ndash44 yearsrsquo olda 10569 54615ndash24 yearsrsquo old 2297 119

Type of airline (moderating variable)Legacy 5884 304Low cost 13475 696

Note aReference categories in the MANOVA model chosen either because they are the largestcategories or because they are the most standard conceptually considered

23Determinants in tourist expenditure composition

Table 3 Frequency distributions of endogenous variables (not included in the MANOVAmodel)

Variables Count Percentage

ActivitiesGolfYes 227 12No 19132 988HikingYes 364 19No 18995 981Sporting eventsYes 603 31No 18756 969Nautical sportsYes 1645 85No 17714 915Other sportsYes 904 47No 18455 953Cultural visitsYes 11180 578No 8179 422Cultural eventsYes 2452 127No 16907 873Other cultural activitiesYes 4175 216No 15184 784SpaYes 1199 62No 18160 938Theme parksYes 1745 90No 17614 910GastronomyYes 1565 81No 17794 919Type of accommodationHotel 4ndash5 5056 261Hotel 3 4922 254Hotel lt 3 6673 345Rented appartment 2116 109Other accommodation 592 31Length of stay5 or less nights 9911 5126ndash8 nights 5952 3079ndash12 nights 1592 8213ndash15 nights 1261 6516 and more nights 643 33

TOURISM ECONOMICS24

Table 3 continued

Variables Count Percentage

Total expenditureQuartile 1 5074 262Quartile 2 5104 264Quartile 3 4986 258Quartile 4 4195 217Expenditure at destination per dayQuartile 1 4855 251Quartile 2 4832 250Quartile 3 4868 251Quartile 4 4804 248Motivation-destinationLeisure on coastin countryside 8083 418Urban tourism 5324 275Others 5952 307

Level of income in contrast affects both log-ratios The higher the incomethe lower the transport share The higher the income the lower the basicdiscretionary expense ratio that is a higher portion of the non-transportexpense is devoted to the discretionary budget

As far as the country of residence is concerned the results show that it affectsmainly the basicdiscretionary log-ratio and exhibits strong airline typemoderating effects LCA users tend to differ depending on their country ofresidence when distributing non-transport expenses into basic versusdiscretionary In general when compared to the UK and Ireland residentcategory (reference category) all other European residents have a largerdiscretionary share More precisely the French Italians and Portuguese have thehighest discretionary share within LCA users and other European countries(mainly central and eastern European) have the highest discretionary sharewithin legacy airline users

Gender has a small effect which is constant for both types of airline Femaletravellers tend to slightly increase the transport share and the share of basicexpenses within the non-transport expenses

The travel group seems to affect both log-ratios and has a moderating effectwith airline type Those tourists who travel in a family group have a highershare in transport expenses when flying with legacy airlines In the second log-ratio travelling as a family with legacy airlines increases the basic expensesshare compared to the share in discretionary Travelling with friends increasesthe share in non-transport expenses for both airline types Besides those whotravel with friends spend more in the discretionary share compared to basicexpenses and those who fly with LCA even more so Finally travelling aloneincreases the basic expenses share compared to the share in discretionaryexpenses for both airline types

When professional status is considered the results show that it mostly affectsthe basicdiscretionary log-ratio and there are some relevant airline type

25Determinants in tourist expenditure composition

Table 4 MANOVA results

y1 log-ratio y2 log-ratiotransportation basicat-destination discretionary

LCA Legacy LCA Legacy

Intercept 0225 0274 2195 1881

Up to high school 0192 0078 0406 0437University 0 0 0 0

Mediumlow income 0073 0289 0037 0518Medium income 0 0 0 0Mediumhigh income ndash0227 ndash0058 ndash0269 ndash0006High income ndash0495 ndash0190 ndash0463 ndash0141

Residence in Portugal 0041 0069 ndash1136 ndash0580Residence in other European Countries 0326 ndash0012 ndash0772 ndash0820Residence in Belgium ndash0127 ndash0159 ndash0667 ndash0461Residence in Netherlands ndash0120 ndash0197 ndash0678 ndash0530Residence in Austria + Switzerland +Liechtenstein ndash0017 ndash0087 ndash0687 ndash0436

Residence in France ndash0112 ndash0111 ndash0870 ndash0553Residence in Scandinavian countries ndash0045 ndash0038 ndash0660 ndash0477Residence in Germany ndash0283 ndash0176 ndash0452 ndash0392Residence in Italy ndash0209 ndash0266 ndash0855 ndash0503Residence in UK or Ireland 0 0 0 0

Female 0094 0094 0093 0093Male 0 0 0 0

Travelling alone ndash0053 ndash0002 0081 0126Travelling in family 0085 0193 0014 0213Travelling with friends ndash0237 ndash0191 ndash0504 ndash0282Travelling with partner 0 0 0 0

Homemaker 0080 0160 0253 0235Unemployed 0030 0159 ndash0036 ndash0039Student ndash0005 ndash0104 ndash0331 ndash0624Low-level employee ndash0034 0194 ndash0194 0063Mid-level employee 0 0 0 0High-level employee ndash0008 ndash0064 ndash0001 ndash0139Self-employed 0029 ndash0086 0193 ndash0072

Over 45 yearsrsquo old and pensioner ndash0060 ndash0060 0633 0633Over 45 yearsrsquo old and not pensioner ndash0010 ndash0010 0142 014225ndash44 yearsrsquo old 0 0 0 015ndash24 yearsrsquo old 0002 0002 ndash0019 ndash0019

moderating effects Legacy users with a low professional status (homemakerunemployed or low-level employee) spend more on transport compared to thenon-transport expenses Conversely students using legacy airlines spend moreon non-transport expenses and these non-transport expenses are to a greater

TOURISM ECONOMICS26

Figure 2 Endogenous variables on the log-ratio space Mean log-ratio valueswithin each category of the endogenous variables

extent devoted to discretionary expenses compared to basic expensesHomemakers increase basic expenditure compared to discretionary with respectto the reference category (mid-level employee) for both airline types LCA userswho are self-employed also increase basic expenditure compared to discretionaryLow-level employees flying with an LCA increase the discretionary sharecompared to the basic and high-level employees flying with legacy airlines alsospend relatively more on the discretionary part

The results show that age affects the second log-ratio more than the firstand there is no moderating effect with the airline type Being a pensioner resultsin a higher share in basic expenses compared to discretionary expenses for bothairline types The remaining categories have hardly any differences from thereference 25ndash44 yearsrsquo old

Results for the endogenous variables

As introduced in the methodology section Figure 2 shows how the endogenousvariables behave in the log-ratio space The origin of the graph is representedby the mean log-ratios in Table 1 01506 for the transportother ratio and17657 for the basicdiscretionary ratio and according to this the graph showscategories with higher or lower ratios than the average

075

060

045

030

015

000

ndash015

ndash030

ndash045

ndash060

Ratio tra

nsport

ationa

t-destination e

xpenses

Ratio basicdiscretionary expenses

050 075 100 125 150 175 200 225 250

27Determinants in tourist expenditure composition

Regarding accommodation it seems that rented apartments and other typesof accommodation behave similarly and have the lowest budget share intransport In the case of hotels as the star rating increases expenditure on basiccompared to discretionary increases and expenditure on transport compared toat-destination decreases

As far as the length of stay is concerned longer stays increase the share ofnon-transport expenditure in basic expenses and reduce the share in transport

As regards destination and motivation tourists going to a seaside or country-side destination for leisure purposes behave quite differently from those goingto an urban destination for urban or cultural tourism The former group spendsmore on transport and less on the discretionary portion However those comingfor urban tourism show an increase in the share of non-transport expenses aswell as in the discretionary share compared to basic expenses

The total expenditure variable gives relevant information It is ordered alongthe first log-ratio axis from top to bottom meaning that those tourists havinga lower total expenditure spend relatively more on the transport share and thosewho have a higher total expenditure spend more on the other two parts of thebudget Regarding the expenditure made at destination per day it is ordereddiagonally from the upper-right corner to the bottom-left corner meaning thattourists with a higher at-destination expenditure per day spend less ontransportation and at the same time more on the discretionary part within at-destination expenses

Finally activities undertaken are located in the lower left quadrant of thegraph meaning that undertaking activities (discretionary expenses) decreasesthe share in transport expenses and although they continue to spend more onbasic expenses than on discretionary the discretionary share increases We havehighlighted some of them Those tourists attending sports and cultural eventsare those spending the most on discretionary compared to basic expensesHiking or some nautical sports could be free of charge thus tourists doing theseactivities have a basic to discretionary ratio close to the mean As for thetransportation share all activities tend to reduce this share about equally withthe exception of golf which leads to a much more substantial reduction

Conclusion and discussion

The main purpose of this article was to study the composition of touristexpenditure and its determinants that is the drivers of the share of touristexpenditure allocated to the different categories of a travel budget and withspecial emphasis on the distinction between legacy versus LCA travellers

The main differences between legacy and LCA users are that income somecountries of residence the travel group and some occupations have differenteffects depending on the type of airline Tourists with a medium or low incomewho travel with legacy airlines tend to spend a greater share of their travelbudget on transportation and basic expenses but those travelling with LCAstend to spend more at the destination and more specifically ondiscretionary expenditure LCA tourists residing in the lsquoother Europeancountriesrsquo spend more on transportation Families flying with legacy airlinesspend more on the transportation share compared to the at-destination share

TOURISM ECONOMICS28

and also spend more on basic expenses compared to discretionary ones Finallyin terms of occupational status legacy users with low-level employment spendmore on transportation whereas the opposite occurs with LCA travellers whospend more on the discretionary component Students and high-level employeesflying with legacy airlines spend more on the discretionary part of the tripbudget

As far as implications for management are concerned assuming that theobjectives of DMOs are to increase the allocation to the at-destinationcomponent of the trip budget (that is lowering the first log-ratio) some of therecommended actions would be to increase marketing efforts directed atmedium- and high-income earners especially at those travelling on LCAs toincrease marketing strategies in some of the European outbound markets suchas Germany (especially those using LCA) and Italy and the Netherlands(especially those flying with legacy airlines) DMOs should also focus on thosetourists travelling with friends (airline type is inconsequential) and studentsusing legacy companies As for the endogenous variables DMOs should directtheir efforts towards those tourists undertaking activities towards those stayinga little bit longer than a week and towards those with high total expenditureFurthermore if DMOs are also interested in capturing the groups who spendrelatively more on discretionary expenses (lowering the second log-ratio) theyshould focus their marketing efforts on those tourists who are highly educatedas well as those who have medium to high incomes and who use LCAsAdditionally they should focus on markets other than the UK and IrelandYoung tourists and tourists who travel with friends (where airline type isinconsequential) low-level employees and students using LCAs as well asstudents and high-level employees flying on legacy airlines should also betargeted In relation to the endogenous variables DMOs must take into accountthose tourists undertaking payable activities those on shorter stays or comingfor urban and cultural tourism or other purposes

The appeal of the CODA methodology for studying tourism budgets lies inthe fact that once the variables have been transformed the researcher can usestandard and well understood statistical models with unbounded error termdistributions while ensuring that the results will be compositionally coherentand that the standard statistical assumptions will hold Conversely results arenot compositionally coherent when we use the MANOVA model on raw sharewithout transforming it into log-ratios We have seen with our data that 84of individuals included in the sample have at least one limit of the 90 shareprediction intervals outside the [01] range

Contrary to the classic and highly restrictive alr and clr transformations theCODA methodology offers the potential to construct tailor-made log-ratios whichare intuitive to interpret and suit the research questions at hand A classificationtree of components is a clear and useful tool in this respect We encourageresearchers to use this approach in further research on tourism expenditure whenbudget share division is fundamental to the researchersrsquo questions

In this study we have encountered two limitations First contrary to studiesthat use macro data in micro data studies the coexistence of full-board halfboard bed and breakfast and accommodation-only tourists makes it impossibleto separate accommodation and food expenditure meaningfully Othersubdivisions (for example discretionary expenditure on activities non-grocery

29Determinants in tourist expenditure composition

shopping and moving around the destination) are technically feasible but canonly increase the percentages of zeros The CODA methodology involves somedegree of amalgamation of components Second even though there are someadvantages to using a database from an official statistics institution as in thiscase (large sample size and scope) the main disadvantage is that the set ofavailable variables cannot be controlled by the researcher A further issue to beaddressed is the time dimension As the EGATUR survey is conducted annuallyfurther research can be done to include a repeat cross-section analysis in orderto capture trends in the effects of predictors

References

Aitchison J (1986) The Statistical Analysis of Compositional Data Monographs on Statistics and AppliedProbability Chapman and Hall London (Reprinted in 2003 with additional material by TheBlackburn Press)

Alegre J Mateo S and Pou LL (2009) lsquoParticipation in tourism consumption and the intensityof participation an analysis of their socio-demographic and economic determinantsrsquo TourismEconomics Vol 15 No 3 pp 531ndash546

Bewley R (1982) lsquoThe generalized addilog demand system applied to Australian time series andcross section datarsquo Australian Economic Papers Vol 21 No 38 pp 177ndash192

Bewley R and Fiebig DG (1988) lsquoEstimation of price elasticities for an international telephonedemand modelrsquo The Journal of Industrial Economics Vol 36 No 4 pp 393ndash409

Cai LA (1998) lsquoAnalysing household food expenditure patterns on trips and vacations a Tobitmodersquo Journal of Hospitality and Tourism Research Vol 22 No 4 pp 338ndash358

Cai LA (1999) lsquoRelationship of household characteristics and lodging expenditure on leisure tripsrsquoJournal of Hospitality and Leisure Marketing Vol 6 No 2 pp 5ndash18

Cai LA Hong G-S and Morrison AM (1995) lsquoHousehold expenditure patterns for tourismproducts and servicesrsquo Journal of Travel and Tourism Marketing Vol 4 No 4 pp 15ndash39

Cannon TF and Ford J (2002) lsquoRelationship of demographic and trip characteristics to visitorspending an analysis of sports travel visitors across timersquo Tourism Economics Vol 8 No 3 pp 263ndash271

Castillo-Manzano JL and Marchena-Goacutemez M (2010) lsquoAnalysis of determinants of airline choiceprofiling the LCC passengerrsquo Applied Economics Letters Vol 18 No 1 pp 49ndash53

Chiou Y-C and Chen Y-H (2010) lsquoFactors influencing the intentions of passengers regarding full ser-vice and low cost carriers a notersquo Journal of Air Transport Management Vol 16 No 4 pp 226ndash228

Coenen M and van Eekeren L (2003) lsquoA study of the demand for domestic tourism by Swedishhouseholds using a two-staged budgeting modelrsquo Scandinavian Journal of Hospitality and TourismVol 3 No 2 pp 114ndash133

Craggs R and Schofield P (2009) lsquoExpenditure-based segmentation and visitor profiling at theQuays in Salford UKrsquo Tourism Economics Vol 15 No 1 pp 243ndash260

Deaton A and Muellbauer J (1980) lsquoAn almost ideal demand systemrsquo American Economic ReviewVol 70 No 3 pp 312ndash326

Dellaert B Borgers A and Timmermans H (1997) lsquoConsumer activity pattern choice devel-opment and test of stage-dependent conjoint choice experimentsrsquo Journal of Retailing and ConsumerServices Vol 4 No 1 pp 25ndash37

Divisekera S (2007) Modelling and Estimation of Tourism Demand Elasticities A Study of TourismExpenditure Allocation in Australia Cooperative Research Centre for Sustainable Tourism (CRCST)Gold Coast

Divisekera S (2009) lsquoEx-post demand for Australian tourism goods and servicesrsquo Tourism EconomicsVol 15 No 1 pp 153ndash180

Divisekera S (2010) lsquoEconomics of leisure and non-leisure tourist demand a study of domesticdemand for Australian tourismrsquo Tourism Economics Vol 16 No 1 pp 117ndash136

Dolnicar S Crouch GI Devinney T Huybers T Louviere JJ and Oppewal H (2008)lsquoTourism and discretionary income allocation Heterogeneity among householdsrsquo TourismManagement Vol 29 No 1 pp 44ndash52

Downward P and Lumsdon L (2000) lsquoThe demand for day-visits an analysis of visitor spendingrsquoTourism Economics Vol 6 No 3 pp 251ndash261

TOURISM ECONOMICS20

For this study we did not consider tourists

bull who have essential zeros in accommodation (tourists who own a house at thedestination or tourists who stay with friends or relatives)

bull who do not decide how much they will spend on certain components(business and study trips)

bull who do not pay for the trip themselves (trips paid for by employers familyfriends prizes offers etc)

bull for whom composition is wholly or partly unobserved (package tourists)

The final sample size was N= 19359From the expenditure variables included in the EGATUR survey database

and used in the model as budget components we first put the amount paidfor transportation (x1) This component has no zeros

Second the amount of money paid for accommodation and food isundistinguishable for full-board half-board or bed and breakfast accommodationTherefore we define a joint accommodation and food component (basic expendi-ture) In this component we include not only the amount paid for consumptionin bars and restaurants but also for buying groceries and everyday products insupermarkets (x2) This component has no zeros

Finally EGATUR provides an aggregated expenditure on activities andshopping (except groceries and everyday products) To this we added theconceptually similar amount paid for moving around the destination (publictransportation andor car rented at the destination) in order to build anactivities and shopping component (x3) This component had 98 zeros whichstresses its discretionary character Zeros were replaced as explained in thestatistical approach subsection

We consider as at-destination (or also called non-transportation) expenses thebasic component (food and accommodation) and the discretionary component(activities and shopping)

Share and log-ratios are described in Table 1 Explanatory variables are thelevel of education income country of residence gender travel groupprofessional status (for pensioners we consider their last professional position)and age The age variable is built as a combination of the original age variableand the variable referring to the individual economic situation This is donein order to have a specific pensioner category taking into account the varyingretirement ages across countries and professions and thus capturing thosetourists who have more free time to undertake a trip regardless of physical ageTable 2 shows their categories and frequencies

As endogenous variables which are not included in the MANOVA modelbut are included in the log-ratio space we include activities undertaken at thedestination accommodation length of stay total expenditure quartilesquartiles of expenditure made at destination per day (basic plus discretionaryexpenses) and a combination of motivation and destination This last variablehas been constructed as a combination of the motivation and the destinationvariable We distinguish between first those tourists travelling for culturaltourism to singular cities second those who come just for leisure either inthe countryside or more commonly at the seaside and third other typologies(Table 3)

21Determinants in tourist expenditure composition

Table 1 Percentage share and log-ratio descriptive statistics

Min Max Mean SD Skewness Kurtosis

Transportation component (x1) 003 9611 2846 1387 0837 0789Basic component (x2) 021 9963 5472 1613 ndash0114 ndash0450Discretionary component (x3) 001 8668 1681 1292 0988 1174Transportationat-destinationlog-ratio (y1) ndash757 548 01506 10702 0714 1404

Basicdiscretionary log-ratio (y2) ndash530 872 17657 18173 1573 2151

Results

Results for the exogenous variables

Table 4 presents the MANOVA results The univariate R2 corrected for thedegrees of freedom is 0045 (first log-ratio) and 0105 (second log-ratio) Themultivariate uncorrected R2 is 1 ndash Λ = 0178 and corrected for the degrees offreedom is 0176 We checked that the results did not change when removingthe four outliers with the highest Cookrsquos distance

Table 4 presents the estimates for the two log-ratios the log-ratio oftransportation over the other two categories (y1) and the log-ratio of basicexpenditure (accommodation and food) over discretionary expenditure (activitiesand shopping) (y2) respectively Because of the moderating effects within eachlog-ratio there are two columns representing the effects when flying with anLCA or with a legacy airline (main and moderating effects have already addedtogether for easier reading) If the LCA and legacy columns are different thereis a significant moderating effect (p-value lt 001)

As we are interpreting log-ratios a positive estimate of a given predictorcategory means that tourists in that category spend more on the numeratorcompared to the denominator than tourists in the reference predictor categoryA negative estimate shows the opposite

The intercept term shows the main effect of company type that is thepredicted log-ratio for each type of airline within the reference category for allvariables (university education medium income resident in the UK or Ireland25ndash44 yearsrsquo old male travelling with partner and mid-level employee) Withinthe reference categories legacy users spend a higher proportion on transportationcompared to other expenses and LCA users spend more on basic expensescompared to discretionary

The level of education seems to have more to do with activities undertakenthan with transportation Results show that a lower level of education resultsin higher expenditure on basic expenses compared to discretionary and is almostequal for users of both types of airline For mainly or only LCA users a lowerlevel of education results in a higher share in transport compared to the othertwo categories If we put it in another way there seems to be a distinct highlyeducated LCA user segment which utilizes the savings in transport toincrease expenditure in non-transportation expenses and even more so indiscretionary

TOURISM ECONOMICS22

Table 2 Frequency distributions of exogenous and moderating variables in the MANOVAmodel

Variables Count Percentage

Level of educationUp to high school 6019 311Universitya 13340 689

Income categoryMediumlow 797 41Mediuma 13186 681Mediumhigh 4376 226High 1000 52

Country of residencePortugal 624 32Other European countries 850 44Belgium 1006 52Netherlands 1069 55Austria Switzerland and Liechtenstein 1122 58France 1326 68Scandinavian countries 1789 92Germany 2363 122Italy 3113 181United Kingdom and Irelanda 6097 315

GenderFemale 8530 441Malea 10829 559

Travel groupAlone 2484 128In family 2980 154With friends 4440 229With partnera 9446 488

Professional statusHomemaker 517 27Unemployed 602 31Student 1432 74Low-level employee 915 47Mid-level employeea 10964 586High-level employee 2138 110Self-employed 2791 144

AgeOver 45 yearsrsquo old and pensioner 1484 77Over 45 yearsrsquo old and not pensioner 5009 25925ndash44 yearsrsquo olda 10569 54615ndash24 yearsrsquo old 2297 119

Type of airline (moderating variable)Legacy 5884 304Low cost 13475 696

Note aReference categories in the MANOVA model chosen either because they are the largestcategories or because they are the most standard conceptually considered

23Determinants in tourist expenditure composition

Table 3 Frequency distributions of endogenous variables (not included in the MANOVAmodel)

Variables Count Percentage

ActivitiesGolfYes 227 12No 19132 988HikingYes 364 19No 18995 981Sporting eventsYes 603 31No 18756 969Nautical sportsYes 1645 85No 17714 915Other sportsYes 904 47No 18455 953Cultural visitsYes 11180 578No 8179 422Cultural eventsYes 2452 127No 16907 873Other cultural activitiesYes 4175 216No 15184 784SpaYes 1199 62No 18160 938Theme parksYes 1745 90No 17614 910GastronomyYes 1565 81No 17794 919Type of accommodationHotel 4ndash5 5056 261Hotel 3 4922 254Hotel lt 3 6673 345Rented appartment 2116 109Other accommodation 592 31Length of stay5 or less nights 9911 5126ndash8 nights 5952 3079ndash12 nights 1592 8213ndash15 nights 1261 6516 and more nights 643 33

TOURISM ECONOMICS24

Table 3 continued

Variables Count Percentage

Total expenditureQuartile 1 5074 262Quartile 2 5104 264Quartile 3 4986 258Quartile 4 4195 217Expenditure at destination per dayQuartile 1 4855 251Quartile 2 4832 250Quartile 3 4868 251Quartile 4 4804 248Motivation-destinationLeisure on coastin countryside 8083 418Urban tourism 5324 275Others 5952 307

Level of income in contrast affects both log-ratios The higher the incomethe lower the transport share The higher the income the lower the basicdiscretionary expense ratio that is a higher portion of the non-transportexpense is devoted to the discretionary budget

As far as the country of residence is concerned the results show that it affectsmainly the basicdiscretionary log-ratio and exhibits strong airline typemoderating effects LCA users tend to differ depending on their country ofresidence when distributing non-transport expenses into basic versusdiscretionary In general when compared to the UK and Ireland residentcategory (reference category) all other European residents have a largerdiscretionary share More precisely the French Italians and Portuguese have thehighest discretionary share within LCA users and other European countries(mainly central and eastern European) have the highest discretionary sharewithin legacy airline users

Gender has a small effect which is constant for both types of airline Femaletravellers tend to slightly increase the transport share and the share of basicexpenses within the non-transport expenses

The travel group seems to affect both log-ratios and has a moderating effectwith airline type Those tourists who travel in a family group have a highershare in transport expenses when flying with legacy airlines In the second log-ratio travelling as a family with legacy airlines increases the basic expensesshare compared to the share in discretionary Travelling with friends increasesthe share in non-transport expenses for both airline types Besides those whotravel with friends spend more in the discretionary share compared to basicexpenses and those who fly with LCA even more so Finally travelling aloneincreases the basic expenses share compared to the share in discretionaryexpenses for both airline types

When professional status is considered the results show that it mostly affectsthe basicdiscretionary log-ratio and there are some relevant airline type

25Determinants in tourist expenditure composition

Table 4 MANOVA results

y1 log-ratio y2 log-ratiotransportation basicat-destination discretionary

LCA Legacy LCA Legacy

Intercept 0225 0274 2195 1881

Up to high school 0192 0078 0406 0437University 0 0 0 0

Mediumlow income 0073 0289 0037 0518Medium income 0 0 0 0Mediumhigh income ndash0227 ndash0058 ndash0269 ndash0006High income ndash0495 ndash0190 ndash0463 ndash0141

Residence in Portugal 0041 0069 ndash1136 ndash0580Residence in other European Countries 0326 ndash0012 ndash0772 ndash0820Residence in Belgium ndash0127 ndash0159 ndash0667 ndash0461Residence in Netherlands ndash0120 ndash0197 ndash0678 ndash0530Residence in Austria + Switzerland +Liechtenstein ndash0017 ndash0087 ndash0687 ndash0436

Residence in France ndash0112 ndash0111 ndash0870 ndash0553Residence in Scandinavian countries ndash0045 ndash0038 ndash0660 ndash0477Residence in Germany ndash0283 ndash0176 ndash0452 ndash0392Residence in Italy ndash0209 ndash0266 ndash0855 ndash0503Residence in UK or Ireland 0 0 0 0

Female 0094 0094 0093 0093Male 0 0 0 0

Travelling alone ndash0053 ndash0002 0081 0126Travelling in family 0085 0193 0014 0213Travelling with friends ndash0237 ndash0191 ndash0504 ndash0282Travelling with partner 0 0 0 0

Homemaker 0080 0160 0253 0235Unemployed 0030 0159 ndash0036 ndash0039Student ndash0005 ndash0104 ndash0331 ndash0624Low-level employee ndash0034 0194 ndash0194 0063Mid-level employee 0 0 0 0High-level employee ndash0008 ndash0064 ndash0001 ndash0139Self-employed 0029 ndash0086 0193 ndash0072

Over 45 yearsrsquo old and pensioner ndash0060 ndash0060 0633 0633Over 45 yearsrsquo old and not pensioner ndash0010 ndash0010 0142 014225ndash44 yearsrsquo old 0 0 0 015ndash24 yearsrsquo old 0002 0002 ndash0019 ndash0019

moderating effects Legacy users with a low professional status (homemakerunemployed or low-level employee) spend more on transport compared to thenon-transport expenses Conversely students using legacy airlines spend moreon non-transport expenses and these non-transport expenses are to a greater

TOURISM ECONOMICS26

Figure 2 Endogenous variables on the log-ratio space Mean log-ratio valueswithin each category of the endogenous variables

extent devoted to discretionary expenses compared to basic expensesHomemakers increase basic expenditure compared to discretionary with respectto the reference category (mid-level employee) for both airline types LCA userswho are self-employed also increase basic expenditure compared to discretionaryLow-level employees flying with an LCA increase the discretionary sharecompared to the basic and high-level employees flying with legacy airlines alsospend relatively more on the discretionary part

The results show that age affects the second log-ratio more than the firstand there is no moderating effect with the airline type Being a pensioner resultsin a higher share in basic expenses compared to discretionary expenses for bothairline types The remaining categories have hardly any differences from thereference 25ndash44 yearsrsquo old

Results for the endogenous variables

As introduced in the methodology section Figure 2 shows how the endogenousvariables behave in the log-ratio space The origin of the graph is representedby the mean log-ratios in Table 1 01506 for the transportother ratio and17657 for the basicdiscretionary ratio and according to this the graph showscategories with higher or lower ratios than the average

075

060

045

030

015

000

ndash015

ndash030

ndash045

ndash060

Ratio tra

nsport

ationa

t-destination e

xpenses

Ratio basicdiscretionary expenses

050 075 100 125 150 175 200 225 250

27Determinants in tourist expenditure composition

Regarding accommodation it seems that rented apartments and other typesof accommodation behave similarly and have the lowest budget share intransport In the case of hotels as the star rating increases expenditure on basiccompared to discretionary increases and expenditure on transport compared toat-destination decreases

As far as the length of stay is concerned longer stays increase the share ofnon-transport expenditure in basic expenses and reduce the share in transport

As regards destination and motivation tourists going to a seaside or country-side destination for leisure purposes behave quite differently from those goingto an urban destination for urban or cultural tourism The former group spendsmore on transport and less on the discretionary portion However those comingfor urban tourism show an increase in the share of non-transport expenses aswell as in the discretionary share compared to basic expenses

The total expenditure variable gives relevant information It is ordered alongthe first log-ratio axis from top to bottom meaning that those tourists havinga lower total expenditure spend relatively more on the transport share and thosewho have a higher total expenditure spend more on the other two parts of thebudget Regarding the expenditure made at destination per day it is ordereddiagonally from the upper-right corner to the bottom-left corner meaning thattourists with a higher at-destination expenditure per day spend less ontransportation and at the same time more on the discretionary part within at-destination expenses

Finally activities undertaken are located in the lower left quadrant of thegraph meaning that undertaking activities (discretionary expenses) decreasesthe share in transport expenses and although they continue to spend more onbasic expenses than on discretionary the discretionary share increases We havehighlighted some of them Those tourists attending sports and cultural eventsare those spending the most on discretionary compared to basic expensesHiking or some nautical sports could be free of charge thus tourists doing theseactivities have a basic to discretionary ratio close to the mean As for thetransportation share all activities tend to reduce this share about equally withthe exception of golf which leads to a much more substantial reduction

Conclusion and discussion

The main purpose of this article was to study the composition of touristexpenditure and its determinants that is the drivers of the share of touristexpenditure allocated to the different categories of a travel budget and withspecial emphasis on the distinction between legacy versus LCA travellers

The main differences between legacy and LCA users are that income somecountries of residence the travel group and some occupations have differenteffects depending on the type of airline Tourists with a medium or low incomewho travel with legacy airlines tend to spend a greater share of their travelbudget on transportation and basic expenses but those travelling with LCAstend to spend more at the destination and more specifically ondiscretionary expenditure LCA tourists residing in the lsquoother Europeancountriesrsquo spend more on transportation Families flying with legacy airlinesspend more on the transportation share compared to the at-destination share

TOURISM ECONOMICS28

and also spend more on basic expenses compared to discretionary ones Finallyin terms of occupational status legacy users with low-level employment spendmore on transportation whereas the opposite occurs with LCA travellers whospend more on the discretionary component Students and high-level employeesflying with legacy airlines spend more on the discretionary part of the tripbudget

As far as implications for management are concerned assuming that theobjectives of DMOs are to increase the allocation to the at-destinationcomponent of the trip budget (that is lowering the first log-ratio) some of therecommended actions would be to increase marketing efforts directed atmedium- and high-income earners especially at those travelling on LCAs toincrease marketing strategies in some of the European outbound markets suchas Germany (especially those using LCA) and Italy and the Netherlands(especially those flying with legacy airlines) DMOs should also focus on thosetourists travelling with friends (airline type is inconsequential) and studentsusing legacy companies As for the endogenous variables DMOs should directtheir efforts towards those tourists undertaking activities towards those stayinga little bit longer than a week and towards those with high total expenditureFurthermore if DMOs are also interested in capturing the groups who spendrelatively more on discretionary expenses (lowering the second log-ratio) theyshould focus their marketing efforts on those tourists who are highly educatedas well as those who have medium to high incomes and who use LCAsAdditionally they should focus on markets other than the UK and IrelandYoung tourists and tourists who travel with friends (where airline type isinconsequential) low-level employees and students using LCAs as well asstudents and high-level employees flying on legacy airlines should also betargeted In relation to the endogenous variables DMOs must take into accountthose tourists undertaking payable activities those on shorter stays or comingfor urban and cultural tourism or other purposes

The appeal of the CODA methodology for studying tourism budgets lies inthe fact that once the variables have been transformed the researcher can usestandard and well understood statistical models with unbounded error termdistributions while ensuring that the results will be compositionally coherentand that the standard statistical assumptions will hold Conversely results arenot compositionally coherent when we use the MANOVA model on raw sharewithout transforming it into log-ratios We have seen with our data that 84of individuals included in the sample have at least one limit of the 90 shareprediction intervals outside the [01] range

Contrary to the classic and highly restrictive alr and clr transformations theCODA methodology offers the potential to construct tailor-made log-ratios whichare intuitive to interpret and suit the research questions at hand A classificationtree of components is a clear and useful tool in this respect We encourageresearchers to use this approach in further research on tourism expenditure whenbudget share division is fundamental to the researchersrsquo questions

In this study we have encountered two limitations First contrary to studiesthat use macro data in micro data studies the coexistence of full-board halfboard bed and breakfast and accommodation-only tourists makes it impossibleto separate accommodation and food expenditure meaningfully Othersubdivisions (for example discretionary expenditure on activities non-grocery

29Determinants in tourist expenditure composition

shopping and moving around the destination) are technically feasible but canonly increase the percentages of zeros The CODA methodology involves somedegree of amalgamation of components Second even though there are someadvantages to using a database from an official statistics institution as in thiscase (large sample size and scope) the main disadvantage is that the set ofavailable variables cannot be controlled by the researcher A further issue to beaddressed is the time dimension As the EGATUR survey is conducted annuallyfurther research can be done to include a repeat cross-section analysis in orderto capture trends in the effects of predictors

References

Aitchison J (1986) The Statistical Analysis of Compositional Data Monographs on Statistics and AppliedProbability Chapman and Hall London (Reprinted in 2003 with additional material by TheBlackburn Press)

Alegre J Mateo S and Pou LL (2009) lsquoParticipation in tourism consumption and the intensityof participation an analysis of their socio-demographic and economic determinantsrsquo TourismEconomics Vol 15 No 3 pp 531ndash546

Bewley R (1982) lsquoThe generalized addilog demand system applied to Australian time series andcross section datarsquo Australian Economic Papers Vol 21 No 38 pp 177ndash192

Bewley R and Fiebig DG (1988) lsquoEstimation of price elasticities for an international telephonedemand modelrsquo The Journal of Industrial Economics Vol 36 No 4 pp 393ndash409

Cai LA (1998) lsquoAnalysing household food expenditure patterns on trips and vacations a Tobitmodersquo Journal of Hospitality and Tourism Research Vol 22 No 4 pp 338ndash358

Cai LA (1999) lsquoRelationship of household characteristics and lodging expenditure on leisure tripsrsquoJournal of Hospitality and Leisure Marketing Vol 6 No 2 pp 5ndash18

Cai LA Hong G-S and Morrison AM (1995) lsquoHousehold expenditure patterns for tourismproducts and servicesrsquo Journal of Travel and Tourism Marketing Vol 4 No 4 pp 15ndash39

Cannon TF and Ford J (2002) lsquoRelationship of demographic and trip characteristics to visitorspending an analysis of sports travel visitors across timersquo Tourism Economics Vol 8 No 3 pp 263ndash271

Castillo-Manzano JL and Marchena-Goacutemez M (2010) lsquoAnalysis of determinants of airline choiceprofiling the LCC passengerrsquo Applied Economics Letters Vol 18 No 1 pp 49ndash53

Chiou Y-C and Chen Y-H (2010) lsquoFactors influencing the intentions of passengers regarding full ser-vice and low cost carriers a notersquo Journal of Air Transport Management Vol 16 No 4 pp 226ndash228

Coenen M and van Eekeren L (2003) lsquoA study of the demand for domestic tourism by Swedishhouseholds using a two-staged budgeting modelrsquo Scandinavian Journal of Hospitality and TourismVol 3 No 2 pp 114ndash133

Craggs R and Schofield P (2009) lsquoExpenditure-based segmentation and visitor profiling at theQuays in Salford UKrsquo Tourism Economics Vol 15 No 1 pp 243ndash260

Deaton A and Muellbauer J (1980) lsquoAn almost ideal demand systemrsquo American Economic ReviewVol 70 No 3 pp 312ndash326

Dellaert B Borgers A and Timmermans H (1997) lsquoConsumer activity pattern choice devel-opment and test of stage-dependent conjoint choice experimentsrsquo Journal of Retailing and ConsumerServices Vol 4 No 1 pp 25ndash37

Divisekera S (2007) Modelling and Estimation of Tourism Demand Elasticities A Study of TourismExpenditure Allocation in Australia Cooperative Research Centre for Sustainable Tourism (CRCST)Gold Coast

Divisekera S (2009) lsquoEx-post demand for Australian tourism goods and servicesrsquo Tourism EconomicsVol 15 No 1 pp 153ndash180

Divisekera S (2010) lsquoEconomics of leisure and non-leisure tourist demand a study of domesticdemand for Australian tourismrsquo Tourism Economics Vol 16 No 1 pp 117ndash136

Dolnicar S Crouch GI Devinney T Huybers T Louviere JJ and Oppewal H (2008)lsquoTourism and discretionary income allocation Heterogeneity among householdsrsquo TourismManagement Vol 29 No 1 pp 44ndash52

Downward P and Lumsdon L (2000) lsquoThe demand for day-visits an analysis of visitor spendingrsquoTourism Economics Vol 6 No 3 pp 251ndash261

21Determinants in tourist expenditure composition

Table 1 Percentage share and log-ratio descriptive statistics

Min Max Mean SD Skewness Kurtosis

Transportation component (x1) 003 9611 2846 1387 0837 0789Basic component (x2) 021 9963 5472 1613 ndash0114 ndash0450Discretionary component (x3) 001 8668 1681 1292 0988 1174Transportationat-destinationlog-ratio (y1) ndash757 548 01506 10702 0714 1404

Basicdiscretionary log-ratio (y2) ndash530 872 17657 18173 1573 2151

Results

Results for the exogenous variables

Table 4 presents the MANOVA results The univariate R2 corrected for thedegrees of freedom is 0045 (first log-ratio) and 0105 (second log-ratio) Themultivariate uncorrected R2 is 1 ndash Λ = 0178 and corrected for the degrees offreedom is 0176 We checked that the results did not change when removingthe four outliers with the highest Cookrsquos distance

Table 4 presents the estimates for the two log-ratios the log-ratio oftransportation over the other two categories (y1) and the log-ratio of basicexpenditure (accommodation and food) over discretionary expenditure (activitiesand shopping) (y2) respectively Because of the moderating effects within eachlog-ratio there are two columns representing the effects when flying with anLCA or with a legacy airline (main and moderating effects have already addedtogether for easier reading) If the LCA and legacy columns are different thereis a significant moderating effect (p-value lt 001)

As we are interpreting log-ratios a positive estimate of a given predictorcategory means that tourists in that category spend more on the numeratorcompared to the denominator than tourists in the reference predictor categoryA negative estimate shows the opposite

The intercept term shows the main effect of company type that is thepredicted log-ratio for each type of airline within the reference category for allvariables (university education medium income resident in the UK or Ireland25ndash44 yearsrsquo old male travelling with partner and mid-level employee) Withinthe reference categories legacy users spend a higher proportion on transportationcompared to other expenses and LCA users spend more on basic expensescompared to discretionary

The level of education seems to have more to do with activities undertakenthan with transportation Results show that a lower level of education resultsin higher expenditure on basic expenses compared to discretionary and is almostequal for users of both types of airline For mainly or only LCA users a lowerlevel of education results in a higher share in transport compared to the othertwo categories If we put it in another way there seems to be a distinct highlyeducated LCA user segment which utilizes the savings in transport toincrease expenditure in non-transportation expenses and even more so indiscretionary

TOURISM ECONOMICS22

Table 2 Frequency distributions of exogenous and moderating variables in the MANOVAmodel

Variables Count Percentage

Level of educationUp to high school 6019 311Universitya 13340 689

Income categoryMediumlow 797 41Mediuma 13186 681Mediumhigh 4376 226High 1000 52

Country of residencePortugal 624 32Other European countries 850 44Belgium 1006 52Netherlands 1069 55Austria Switzerland and Liechtenstein 1122 58France 1326 68Scandinavian countries 1789 92Germany 2363 122Italy 3113 181United Kingdom and Irelanda 6097 315

GenderFemale 8530 441Malea 10829 559

Travel groupAlone 2484 128In family 2980 154With friends 4440 229With partnera 9446 488

Professional statusHomemaker 517 27Unemployed 602 31Student 1432 74Low-level employee 915 47Mid-level employeea 10964 586High-level employee 2138 110Self-employed 2791 144

AgeOver 45 yearsrsquo old and pensioner 1484 77Over 45 yearsrsquo old and not pensioner 5009 25925ndash44 yearsrsquo olda 10569 54615ndash24 yearsrsquo old 2297 119

Type of airline (moderating variable)Legacy 5884 304Low cost 13475 696

Note aReference categories in the MANOVA model chosen either because they are the largestcategories or because they are the most standard conceptually considered

23Determinants in tourist expenditure composition

Table 3 Frequency distributions of endogenous variables (not included in the MANOVAmodel)

Variables Count Percentage

ActivitiesGolfYes 227 12No 19132 988HikingYes 364 19No 18995 981Sporting eventsYes 603 31No 18756 969Nautical sportsYes 1645 85No 17714 915Other sportsYes 904 47No 18455 953Cultural visitsYes 11180 578No 8179 422Cultural eventsYes 2452 127No 16907 873Other cultural activitiesYes 4175 216No 15184 784SpaYes 1199 62No 18160 938Theme parksYes 1745 90No 17614 910GastronomyYes 1565 81No 17794 919Type of accommodationHotel 4ndash5 5056 261Hotel 3 4922 254Hotel lt 3 6673 345Rented appartment 2116 109Other accommodation 592 31Length of stay5 or less nights 9911 5126ndash8 nights 5952 3079ndash12 nights 1592 8213ndash15 nights 1261 6516 and more nights 643 33

TOURISM ECONOMICS24

Table 3 continued

Variables Count Percentage

Total expenditureQuartile 1 5074 262Quartile 2 5104 264Quartile 3 4986 258Quartile 4 4195 217Expenditure at destination per dayQuartile 1 4855 251Quartile 2 4832 250Quartile 3 4868 251Quartile 4 4804 248Motivation-destinationLeisure on coastin countryside 8083 418Urban tourism 5324 275Others 5952 307

Level of income in contrast affects both log-ratios The higher the incomethe lower the transport share The higher the income the lower the basicdiscretionary expense ratio that is a higher portion of the non-transportexpense is devoted to the discretionary budget

As far as the country of residence is concerned the results show that it affectsmainly the basicdiscretionary log-ratio and exhibits strong airline typemoderating effects LCA users tend to differ depending on their country ofresidence when distributing non-transport expenses into basic versusdiscretionary In general when compared to the UK and Ireland residentcategory (reference category) all other European residents have a largerdiscretionary share More precisely the French Italians and Portuguese have thehighest discretionary share within LCA users and other European countries(mainly central and eastern European) have the highest discretionary sharewithin legacy airline users

Gender has a small effect which is constant for both types of airline Femaletravellers tend to slightly increase the transport share and the share of basicexpenses within the non-transport expenses

The travel group seems to affect both log-ratios and has a moderating effectwith airline type Those tourists who travel in a family group have a highershare in transport expenses when flying with legacy airlines In the second log-ratio travelling as a family with legacy airlines increases the basic expensesshare compared to the share in discretionary Travelling with friends increasesthe share in non-transport expenses for both airline types Besides those whotravel with friends spend more in the discretionary share compared to basicexpenses and those who fly with LCA even more so Finally travelling aloneincreases the basic expenses share compared to the share in discretionaryexpenses for both airline types

When professional status is considered the results show that it mostly affectsthe basicdiscretionary log-ratio and there are some relevant airline type

25Determinants in tourist expenditure composition

Table 4 MANOVA results

y1 log-ratio y2 log-ratiotransportation basicat-destination discretionary

LCA Legacy LCA Legacy

Intercept 0225 0274 2195 1881

Up to high school 0192 0078 0406 0437University 0 0 0 0

Mediumlow income 0073 0289 0037 0518Medium income 0 0 0 0Mediumhigh income ndash0227 ndash0058 ndash0269 ndash0006High income ndash0495 ndash0190 ndash0463 ndash0141

Residence in Portugal 0041 0069 ndash1136 ndash0580Residence in other European Countries 0326 ndash0012 ndash0772 ndash0820Residence in Belgium ndash0127 ndash0159 ndash0667 ndash0461Residence in Netherlands ndash0120 ndash0197 ndash0678 ndash0530Residence in Austria + Switzerland +Liechtenstein ndash0017 ndash0087 ndash0687 ndash0436

Residence in France ndash0112 ndash0111 ndash0870 ndash0553Residence in Scandinavian countries ndash0045 ndash0038 ndash0660 ndash0477Residence in Germany ndash0283 ndash0176 ndash0452 ndash0392Residence in Italy ndash0209 ndash0266 ndash0855 ndash0503Residence in UK or Ireland 0 0 0 0

Female 0094 0094 0093 0093Male 0 0 0 0

Travelling alone ndash0053 ndash0002 0081 0126Travelling in family 0085 0193 0014 0213Travelling with friends ndash0237 ndash0191 ndash0504 ndash0282Travelling with partner 0 0 0 0

Homemaker 0080 0160 0253 0235Unemployed 0030 0159 ndash0036 ndash0039Student ndash0005 ndash0104 ndash0331 ndash0624Low-level employee ndash0034 0194 ndash0194 0063Mid-level employee 0 0 0 0High-level employee ndash0008 ndash0064 ndash0001 ndash0139Self-employed 0029 ndash0086 0193 ndash0072

Over 45 yearsrsquo old and pensioner ndash0060 ndash0060 0633 0633Over 45 yearsrsquo old and not pensioner ndash0010 ndash0010 0142 014225ndash44 yearsrsquo old 0 0 0 015ndash24 yearsrsquo old 0002 0002 ndash0019 ndash0019

moderating effects Legacy users with a low professional status (homemakerunemployed or low-level employee) spend more on transport compared to thenon-transport expenses Conversely students using legacy airlines spend moreon non-transport expenses and these non-transport expenses are to a greater

TOURISM ECONOMICS26

Figure 2 Endogenous variables on the log-ratio space Mean log-ratio valueswithin each category of the endogenous variables

extent devoted to discretionary expenses compared to basic expensesHomemakers increase basic expenditure compared to discretionary with respectto the reference category (mid-level employee) for both airline types LCA userswho are self-employed also increase basic expenditure compared to discretionaryLow-level employees flying with an LCA increase the discretionary sharecompared to the basic and high-level employees flying with legacy airlines alsospend relatively more on the discretionary part

The results show that age affects the second log-ratio more than the firstand there is no moderating effect with the airline type Being a pensioner resultsin a higher share in basic expenses compared to discretionary expenses for bothairline types The remaining categories have hardly any differences from thereference 25ndash44 yearsrsquo old

Results for the endogenous variables

As introduced in the methodology section Figure 2 shows how the endogenousvariables behave in the log-ratio space The origin of the graph is representedby the mean log-ratios in Table 1 01506 for the transportother ratio and17657 for the basicdiscretionary ratio and according to this the graph showscategories with higher or lower ratios than the average

075

060

045

030

015

000

ndash015

ndash030

ndash045

ndash060

Ratio tra

nsport

ationa

t-destination e

xpenses

Ratio basicdiscretionary expenses

050 075 100 125 150 175 200 225 250

27Determinants in tourist expenditure composition

Regarding accommodation it seems that rented apartments and other typesof accommodation behave similarly and have the lowest budget share intransport In the case of hotels as the star rating increases expenditure on basiccompared to discretionary increases and expenditure on transport compared toat-destination decreases

As far as the length of stay is concerned longer stays increase the share ofnon-transport expenditure in basic expenses and reduce the share in transport

As regards destination and motivation tourists going to a seaside or country-side destination for leisure purposes behave quite differently from those goingto an urban destination for urban or cultural tourism The former group spendsmore on transport and less on the discretionary portion However those comingfor urban tourism show an increase in the share of non-transport expenses aswell as in the discretionary share compared to basic expenses

The total expenditure variable gives relevant information It is ordered alongthe first log-ratio axis from top to bottom meaning that those tourists havinga lower total expenditure spend relatively more on the transport share and thosewho have a higher total expenditure spend more on the other two parts of thebudget Regarding the expenditure made at destination per day it is ordereddiagonally from the upper-right corner to the bottom-left corner meaning thattourists with a higher at-destination expenditure per day spend less ontransportation and at the same time more on the discretionary part within at-destination expenses

Finally activities undertaken are located in the lower left quadrant of thegraph meaning that undertaking activities (discretionary expenses) decreasesthe share in transport expenses and although they continue to spend more onbasic expenses than on discretionary the discretionary share increases We havehighlighted some of them Those tourists attending sports and cultural eventsare those spending the most on discretionary compared to basic expensesHiking or some nautical sports could be free of charge thus tourists doing theseactivities have a basic to discretionary ratio close to the mean As for thetransportation share all activities tend to reduce this share about equally withthe exception of golf which leads to a much more substantial reduction

Conclusion and discussion

The main purpose of this article was to study the composition of touristexpenditure and its determinants that is the drivers of the share of touristexpenditure allocated to the different categories of a travel budget and withspecial emphasis on the distinction between legacy versus LCA travellers

The main differences between legacy and LCA users are that income somecountries of residence the travel group and some occupations have differenteffects depending on the type of airline Tourists with a medium or low incomewho travel with legacy airlines tend to spend a greater share of their travelbudget on transportation and basic expenses but those travelling with LCAstend to spend more at the destination and more specifically ondiscretionary expenditure LCA tourists residing in the lsquoother Europeancountriesrsquo spend more on transportation Families flying with legacy airlinesspend more on the transportation share compared to the at-destination share

TOURISM ECONOMICS28

and also spend more on basic expenses compared to discretionary ones Finallyin terms of occupational status legacy users with low-level employment spendmore on transportation whereas the opposite occurs with LCA travellers whospend more on the discretionary component Students and high-level employeesflying with legacy airlines spend more on the discretionary part of the tripbudget

As far as implications for management are concerned assuming that theobjectives of DMOs are to increase the allocation to the at-destinationcomponent of the trip budget (that is lowering the first log-ratio) some of therecommended actions would be to increase marketing efforts directed atmedium- and high-income earners especially at those travelling on LCAs toincrease marketing strategies in some of the European outbound markets suchas Germany (especially those using LCA) and Italy and the Netherlands(especially those flying with legacy airlines) DMOs should also focus on thosetourists travelling with friends (airline type is inconsequential) and studentsusing legacy companies As for the endogenous variables DMOs should directtheir efforts towards those tourists undertaking activities towards those stayinga little bit longer than a week and towards those with high total expenditureFurthermore if DMOs are also interested in capturing the groups who spendrelatively more on discretionary expenses (lowering the second log-ratio) theyshould focus their marketing efforts on those tourists who are highly educatedas well as those who have medium to high incomes and who use LCAsAdditionally they should focus on markets other than the UK and IrelandYoung tourists and tourists who travel with friends (where airline type isinconsequential) low-level employees and students using LCAs as well asstudents and high-level employees flying on legacy airlines should also betargeted In relation to the endogenous variables DMOs must take into accountthose tourists undertaking payable activities those on shorter stays or comingfor urban and cultural tourism or other purposes

The appeal of the CODA methodology for studying tourism budgets lies inthe fact that once the variables have been transformed the researcher can usestandard and well understood statistical models with unbounded error termdistributions while ensuring that the results will be compositionally coherentand that the standard statistical assumptions will hold Conversely results arenot compositionally coherent when we use the MANOVA model on raw sharewithout transforming it into log-ratios We have seen with our data that 84of individuals included in the sample have at least one limit of the 90 shareprediction intervals outside the [01] range

Contrary to the classic and highly restrictive alr and clr transformations theCODA methodology offers the potential to construct tailor-made log-ratios whichare intuitive to interpret and suit the research questions at hand A classificationtree of components is a clear and useful tool in this respect We encourageresearchers to use this approach in further research on tourism expenditure whenbudget share division is fundamental to the researchersrsquo questions

In this study we have encountered two limitations First contrary to studiesthat use macro data in micro data studies the coexistence of full-board halfboard bed and breakfast and accommodation-only tourists makes it impossibleto separate accommodation and food expenditure meaningfully Othersubdivisions (for example discretionary expenditure on activities non-grocery

29Determinants in tourist expenditure composition

shopping and moving around the destination) are technically feasible but canonly increase the percentages of zeros The CODA methodology involves somedegree of amalgamation of components Second even though there are someadvantages to using a database from an official statistics institution as in thiscase (large sample size and scope) the main disadvantage is that the set ofavailable variables cannot be controlled by the researcher A further issue to beaddressed is the time dimension As the EGATUR survey is conducted annuallyfurther research can be done to include a repeat cross-section analysis in orderto capture trends in the effects of predictors

References

Aitchison J (1986) The Statistical Analysis of Compositional Data Monographs on Statistics and AppliedProbability Chapman and Hall London (Reprinted in 2003 with additional material by TheBlackburn Press)

Alegre J Mateo S and Pou LL (2009) lsquoParticipation in tourism consumption and the intensityof participation an analysis of their socio-demographic and economic determinantsrsquo TourismEconomics Vol 15 No 3 pp 531ndash546

Bewley R (1982) lsquoThe generalized addilog demand system applied to Australian time series andcross section datarsquo Australian Economic Papers Vol 21 No 38 pp 177ndash192

Bewley R and Fiebig DG (1988) lsquoEstimation of price elasticities for an international telephonedemand modelrsquo The Journal of Industrial Economics Vol 36 No 4 pp 393ndash409

Cai LA (1998) lsquoAnalysing household food expenditure patterns on trips and vacations a Tobitmodersquo Journal of Hospitality and Tourism Research Vol 22 No 4 pp 338ndash358

Cai LA (1999) lsquoRelationship of household characteristics and lodging expenditure on leisure tripsrsquoJournal of Hospitality and Leisure Marketing Vol 6 No 2 pp 5ndash18

Cai LA Hong G-S and Morrison AM (1995) lsquoHousehold expenditure patterns for tourismproducts and servicesrsquo Journal of Travel and Tourism Marketing Vol 4 No 4 pp 15ndash39

Cannon TF and Ford J (2002) lsquoRelationship of demographic and trip characteristics to visitorspending an analysis of sports travel visitors across timersquo Tourism Economics Vol 8 No 3 pp 263ndash271

Castillo-Manzano JL and Marchena-Goacutemez M (2010) lsquoAnalysis of determinants of airline choiceprofiling the LCC passengerrsquo Applied Economics Letters Vol 18 No 1 pp 49ndash53

Chiou Y-C and Chen Y-H (2010) lsquoFactors influencing the intentions of passengers regarding full ser-vice and low cost carriers a notersquo Journal of Air Transport Management Vol 16 No 4 pp 226ndash228

Coenen M and van Eekeren L (2003) lsquoA study of the demand for domestic tourism by Swedishhouseholds using a two-staged budgeting modelrsquo Scandinavian Journal of Hospitality and TourismVol 3 No 2 pp 114ndash133

Craggs R and Schofield P (2009) lsquoExpenditure-based segmentation and visitor profiling at theQuays in Salford UKrsquo Tourism Economics Vol 15 No 1 pp 243ndash260

Deaton A and Muellbauer J (1980) lsquoAn almost ideal demand systemrsquo American Economic ReviewVol 70 No 3 pp 312ndash326

Dellaert B Borgers A and Timmermans H (1997) lsquoConsumer activity pattern choice devel-opment and test of stage-dependent conjoint choice experimentsrsquo Journal of Retailing and ConsumerServices Vol 4 No 1 pp 25ndash37

Divisekera S (2007) Modelling and Estimation of Tourism Demand Elasticities A Study of TourismExpenditure Allocation in Australia Cooperative Research Centre for Sustainable Tourism (CRCST)Gold Coast

Divisekera S (2009) lsquoEx-post demand for Australian tourism goods and servicesrsquo Tourism EconomicsVol 15 No 1 pp 153ndash180

Divisekera S (2010) lsquoEconomics of leisure and non-leisure tourist demand a study of domesticdemand for Australian tourismrsquo Tourism Economics Vol 16 No 1 pp 117ndash136

Dolnicar S Crouch GI Devinney T Huybers T Louviere JJ and Oppewal H (2008)lsquoTourism and discretionary income allocation Heterogeneity among householdsrsquo TourismManagement Vol 29 No 1 pp 44ndash52

Downward P and Lumsdon L (2000) lsquoThe demand for day-visits an analysis of visitor spendingrsquoTourism Economics Vol 6 No 3 pp 251ndash261

TOURISM ECONOMICS22

Table 2 Frequency distributions of exogenous and moderating variables in the MANOVAmodel

Variables Count Percentage

Level of educationUp to high school 6019 311Universitya 13340 689

Income categoryMediumlow 797 41Mediuma 13186 681Mediumhigh 4376 226High 1000 52

Country of residencePortugal 624 32Other European countries 850 44Belgium 1006 52Netherlands 1069 55Austria Switzerland and Liechtenstein 1122 58France 1326 68Scandinavian countries 1789 92Germany 2363 122Italy 3113 181United Kingdom and Irelanda 6097 315

GenderFemale 8530 441Malea 10829 559

Travel groupAlone 2484 128In family 2980 154With friends 4440 229With partnera 9446 488

Professional statusHomemaker 517 27Unemployed 602 31Student 1432 74Low-level employee 915 47Mid-level employeea 10964 586High-level employee 2138 110Self-employed 2791 144

AgeOver 45 yearsrsquo old and pensioner 1484 77Over 45 yearsrsquo old and not pensioner 5009 25925ndash44 yearsrsquo olda 10569 54615ndash24 yearsrsquo old 2297 119

Type of airline (moderating variable)Legacy 5884 304Low cost 13475 696

Note aReference categories in the MANOVA model chosen either because they are the largestcategories or because they are the most standard conceptually considered

23Determinants in tourist expenditure composition

Table 3 Frequency distributions of endogenous variables (not included in the MANOVAmodel)

Variables Count Percentage

ActivitiesGolfYes 227 12No 19132 988HikingYes 364 19No 18995 981Sporting eventsYes 603 31No 18756 969Nautical sportsYes 1645 85No 17714 915Other sportsYes 904 47No 18455 953Cultural visitsYes 11180 578No 8179 422Cultural eventsYes 2452 127No 16907 873Other cultural activitiesYes 4175 216No 15184 784SpaYes 1199 62No 18160 938Theme parksYes 1745 90No 17614 910GastronomyYes 1565 81No 17794 919Type of accommodationHotel 4ndash5 5056 261Hotel 3 4922 254Hotel lt 3 6673 345Rented appartment 2116 109Other accommodation 592 31Length of stay5 or less nights 9911 5126ndash8 nights 5952 3079ndash12 nights 1592 8213ndash15 nights 1261 6516 and more nights 643 33

TOURISM ECONOMICS24

Table 3 continued

Variables Count Percentage

Total expenditureQuartile 1 5074 262Quartile 2 5104 264Quartile 3 4986 258Quartile 4 4195 217Expenditure at destination per dayQuartile 1 4855 251Quartile 2 4832 250Quartile 3 4868 251Quartile 4 4804 248Motivation-destinationLeisure on coastin countryside 8083 418Urban tourism 5324 275Others 5952 307

Level of income in contrast affects both log-ratios The higher the incomethe lower the transport share The higher the income the lower the basicdiscretionary expense ratio that is a higher portion of the non-transportexpense is devoted to the discretionary budget

As far as the country of residence is concerned the results show that it affectsmainly the basicdiscretionary log-ratio and exhibits strong airline typemoderating effects LCA users tend to differ depending on their country ofresidence when distributing non-transport expenses into basic versusdiscretionary In general when compared to the UK and Ireland residentcategory (reference category) all other European residents have a largerdiscretionary share More precisely the French Italians and Portuguese have thehighest discretionary share within LCA users and other European countries(mainly central and eastern European) have the highest discretionary sharewithin legacy airline users

Gender has a small effect which is constant for both types of airline Femaletravellers tend to slightly increase the transport share and the share of basicexpenses within the non-transport expenses

The travel group seems to affect both log-ratios and has a moderating effectwith airline type Those tourists who travel in a family group have a highershare in transport expenses when flying with legacy airlines In the second log-ratio travelling as a family with legacy airlines increases the basic expensesshare compared to the share in discretionary Travelling with friends increasesthe share in non-transport expenses for both airline types Besides those whotravel with friends spend more in the discretionary share compared to basicexpenses and those who fly with LCA even more so Finally travelling aloneincreases the basic expenses share compared to the share in discretionaryexpenses for both airline types

When professional status is considered the results show that it mostly affectsthe basicdiscretionary log-ratio and there are some relevant airline type

25Determinants in tourist expenditure composition

Table 4 MANOVA results

y1 log-ratio y2 log-ratiotransportation basicat-destination discretionary

LCA Legacy LCA Legacy

Intercept 0225 0274 2195 1881

Up to high school 0192 0078 0406 0437University 0 0 0 0

Mediumlow income 0073 0289 0037 0518Medium income 0 0 0 0Mediumhigh income ndash0227 ndash0058 ndash0269 ndash0006High income ndash0495 ndash0190 ndash0463 ndash0141

Residence in Portugal 0041 0069 ndash1136 ndash0580Residence in other European Countries 0326 ndash0012 ndash0772 ndash0820Residence in Belgium ndash0127 ndash0159 ndash0667 ndash0461Residence in Netherlands ndash0120 ndash0197 ndash0678 ndash0530Residence in Austria + Switzerland +Liechtenstein ndash0017 ndash0087 ndash0687 ndash0436

Residence in France ndash0112 ndash0111 ndash0870 ndash0553Residence in Scandinavian countries ndash0045 ndash0038 ndash0660 ndash0477Residence in Germany ndash0283 ndash0176 ndash0452 ndash0392Residence in Italy ndash0209 ndash0266 ndash0855 ndash0503Residence in UK or Ireland 0 0 0 0

Female 0094 0094 0093 0093Male 0 0 0 0

Travelling alone ndash0053 ndash0002 0081 0126Travelling in family 0085 0193 0014 0213Travelling with friends ndash0237 ndash0191 ndash0504 ndash0282Travelling with partner 0 0 0 0

Homemaker 0080 0160 0253 0235Unemployed 0030 0159 ndash0036 ndash0039Student ndash0005 ndash0104 ndash0331 ndash0624Low-level employee ndash0034 0194 ndash0194 0063Mid-level employee 0 0 0 0High-level employee ndash0008 ndash0064 ndash0001 ndash0139Self-employed 0029 ndash0086 0193 ndash0072

Over 45 yearsrsquo old and pensioner ndash0060 ndash0060 0633 0633Over 45 yearsrsquo old and not pensioner ndash0010 ndash0010 0142 014225ndash44 yearsrsquo old 0 0 0 015ndash24 yearsrsquo old 0002 0002 ndash0019 ndash0019

moderating effects Legacy users with a low professional status (homemakerunemployed or low-level employee) spend more on transport compared to thenon-transport expenses Conversely students using legacy airlines spend moreon non-transport expenses and these non-transport expenses are to a greater

TOURISM ECONOMICS26

Figure 2 Endogenous variables on the log-ratio space Mean log-ratio valueswithin each category of the endogenous variables

extent devoted to discretionary expenses compared to basic expensesHomemakers increase basic expenditure compared to discretionary with respectto the reference category (mid-level employee) for both airline types LCA userswho are self-employed also increase basic expenditure compared to discretionaryLow-level employees flying with an LCA increase the discretionary sharecompared to the basic and high-level employees flying with legacy airlines alsospend relatively more on the discretionary part

The results show that age affects the second log-ratio more than the firstand there is no moderating effect with the airline type Being a pensioner resultsin a higher share in basic expenses compared to discretionary expenses for bothairline types The remaining categories have hardly any differences from thereference 25ndash44 yearsrsquo old

Results for the endogenous variables

As introduced in the methodology section Figure 2 shows how the endogenousvariables behave in the log-ratio space The origin of the graph is representedby the mean log-ratios in Table 1 01506 for the transportother ratio and17657 for the basicdiscretionary ratio and according to this the graph showscategories with higher or lower ratios than the average

075

060

045

030

015

000

ndash015

ndash030

ndash045

ndash060

Ratio tra

nsport

ationa

t-destination e

xpenses

Ratio basicdiscretionary expenses

050 075 100 125 150 175 200 225 250

27Determinants in tourist expenditure composition

Regarding accommodation it seems that rented apartments and other typesof accommodation behave similarly and have the lowest budget share intransport In the case of hotels as the star rating increases expenditure on basiccompared to discretionary increases and expenditure on transport compared toat-destination decreases

As far as the length of stay is concerned longer stays increase the share ofnon-transport expenditure in basic expenses and reduce the share in transport

As regards destination and motivation tourists going to a seaside or country-side destination for leisure purposes behave quite differently from those goingto an urban destination for urban or cultural tourism The former group spendsmore on transport and less on the discretionary portion However those comingfor urban tourism show an increase in the share of non-transport expenses aswell as in the discretionary share compared to basic expenses

The total expenditure variable gives relevant information It is ordered alongthe first log-ratio axis from top to bottom meaning that those tourists havinga lower total expenditure spend relatively more on the transport share and thosewho have a higher total expenditure spend more on the other two parts of thebudget Regarding the expenditure made at destination per day it is ordereddiagonally from the upper-right corner to the bottom-left corner meaning thattourists with a higher at-destination expenditure per day spend less ontransportation and at the same time more on the discretionary part within at-destination expenses

Finally activities undertaken are located in the lower left quadrant of thegraph meaning that undertaking activities (discretionary expenses) decreasesthe share in transport expenses and although they continue to spend more onbasic expenses than on discretionary the discretionary share increases We havehighlighted some of them Those tourists attending sports and cultural eventsare those spending the most on discretionary compared to basic expensesHiking or some nautical sports could be free of charge thus tourists doing theseactivities have a basic to discretionary ratio close to the mean As for thetransportation share all activities tend to reduce this share about equally withthe exception of golf which leads to a much more substantial reduction

Conclusion and discussion

The main purpose of this article was to study the composition of touristexpenditure and its determinants that is the drivers of the share of touristexpenditure allocated to the different categories of a travel budget and withspecial emphasis on the distinction between legacy versus LCA travellers

The main differences between legacy and LCA users are that income somecountries of residence the travel group and some occupations have differenteffects depending on the type of airline Tourists with a medium or low incomewho travel with legacy airlines tend to spend a greater share of their travelbudget on transportation and basic expenses but those travelling with LCAstend to spend more at the destination and more specifically ondiscretionary expenditure LCA tourists residing in the lsquoother Europeancountriesrsquo spend more on transportation Families flying with legacy airlinesspend more on the transportation share compared to the at-destination share

TOURISM ECONOMICS28

and also spend more on basic expenses compared to discretionary ones Finallyin terms of occupational status legacy users with low-level employment spendmore on transportation whereas the opposite occurs with LCA travellers whospend more on the discretionary component Students and high-level employeesflying with legacy airlines spend more on the discretionary part of the tripbudget

As far as implications for management are concerned assuming that theobjectives of DMOs are to increase the allocation to the at-destinationcomponent of the trip budget (that is lowering the first log-ratio) some of therecommended actions would be to increase marketing efforts directed atmedium- and high-income earners especially at those travelling on LCAs toincrease marketing strategies in some of the European outbound markets suchas Germany (especially those using LCA) and Italy and the Netherlands(especially those flying with legacy airlines) DMOs should also focus on thosetourists travelling with friends (airline type is inconsequential) and studentsusing legacy companies As for the endogenous variables DMOs should directtheir efforts towards those tourists undertaking activities towards those stayinga little bit longer than a week and towards those with high total expenditureFurthermore if DMOs are also interested in capturing the groups who spendrelatively more on discretionary expenses (lowering the second log-ratio) theyshould focus their marketing efforts on those tourists who are highly educatedas well as those who have medium to high incomes and who use LCAsAdditionally they should focus on markets other than the UK and IrelandYoung tourists and tourists who travel with friends (where airline type isinconsequential) low-level employees and students using LCAs as well asstudents and high-level employees flying on legacy airlines should also betargeted In relation to the endogenous variables DMOs must take into accountthose tourists undertaking payable activities those on shorter stays or comingfor urban and cultural tourism or other purposes

The appeal of the CODA methodology for studying tourism budgets lies inthe fact that once the variables have been transformed the researcher can usestandard and well understood statistical models with unbounded error termdistributions while ensuring that the results will be compositionally coherentand that the standard statistical assumptions will hold Conversely results arenot compositionally coherent when we use the MANOVA model on raw sharewithout transforming it into log-ratios We have seen with our data that 84of individuals included in the sample have at least one limit of the 90 shareprediction intervals outside the [01] range

Contrary to the classic and highly restrictive alr and clr transformations theCODA methodology offers the potential to construct tailor-made log-ratios whichare intuitive to interpret and suit the research questions at hand A classificationtree of components is a clear and useful tool in this respect We encourageresearchers to use this approach in further research on tourism expenditure whenbudget share division is fundamental to the researchersrsquo questions

In this study we have encountered two limitations First contrary to studiesthat use macro data in micro data studies the coexistence of full-board halfboard bed and breakfast and accommodation-only tourists makes it impossibleto separate accommodation and food expenditure meaningfully Othersubdivisions (for example discretionary expenditure on activities non-grocery

29Determinants in tourist expenditure composition

shopping and moving around the destination) are technically feasible but canonly increase the percentages of zeros The CODA methodology involves somedegree of amalgamation of components Second even though there are someadvantages to using a database from an official statistics institution as in thiscase (large sample size and scope) the main disadvantage is that the set ofavailable variables cannot be controlled by the researcher A further issue to beaddressed is the time dimension As the EGATUR survey is conducted annuallyfurther research can be done to include a repeat cross-section analysis in orderto capture trends in the effects of predictors

References

Aitchison J (1986) The Statistical Analysis of Compositional Data Monographs on Statistics and AppliedProbability Chapman and Hall London (Reprinted in 2003 with additional material by TheBlackburn Press)

Alegre J Mateo S and Pou LL (2009) lsquoParticipation in tourism consumption and the intensityof participation an analysis of their socio-demographic and economic determinantsrsquo TourismEconomics Vol 15 No 3 pp 531ndash546

Bewley R (1982) lsquoThe generalized addilog demand system applied to Australian time series andcross section datarsquo Australian Economic Papers Vol 21 No 38 pp 177ndash192

Bewley R and Fiebig DG (1988) lsquoEstimation of price elasticities for an international telephonedemand modelrsquo The Journal of Industrial Economics Vol 36 No 4 pp 393ndash409

Cai LA (1998) lsquoAnalysing household food expenditure patterns on trips and vacations a Tobitmodersquo Journal of Hospitality and Tourism Research Vol 22 No 4 pp 338ndash358

Cai LA (1999) lsquoRelationship of household characteristics and lodging expenditure on leisure tripsrsquoJournal of Hospitality and Leisure Marketing Vol 6 No 2 pp 5ndash18

Cai LA Hong G-S and Morrison AM (1995) lsquoHousehold expenditure patterns for tourismproducts and servicesrsquo Journal of Travel and Tourism Marketing Vol 4 No 4 pp 15ndash39

Cannon TF and Ford J (2002) lsquoRelationship of demographic and trip characteristics to visitorspending an analysis of sports travel visitors across timersquo Tourism Economics Vol 8 No 3 pp 263ndash271

Castillo-Manzano JL and Marchena-Goacutemez M (2010) lsquoAnalysis of determinants of airline choiceprofiling the LCC passengerrsquo Applied Economics Letters Vol 18 No 1 pp 49ndash53

Chiou Y-C and Chen Y-H (2010) lsquoFactors influencing the intentions of passengers regarding full ser-vice and low cost carriers a notersquo Journal of Air Transport Management Vol 16 No 4 pp 226ndash228

Coenen M and van Eekeren L (2003) lsquoA study of the demand for domestic tourism by Swedishhouseholds using a two-staged budgeting modelrsquo Scandinavian Journal of Hospitality and TourismVol 3 No 2 pp 114ndash133

Craggs R and Schofield P (2009) lsquoExpenditure-based segmentation and visitor profiling at theQuays in Salford UKrsquo Tourism Economics Vol 15 No 1 pp 243ndash260

Deaton A and Muellbauer J (1980) lsquoAn almost ideal demand systemrsquo American Economic ReviewVol 70 No 3 pp 312ndash326

Dellaert B Borgers A and Timmermans H (1997) lsquoConsumer activity pattern choice devel-opment and test of stage-dependent conjoint choice experimentsrsquo Journal of Retailing and ConsumerServices Vol 4 No 1 pp 25ndash37

Divisekera S (2007) Modelling and Estimation of Tourism Demand Elasticities A Study of TourismExpenditure Allocation in Australia Cooperative Research Centre for Sustainable Tourism (CRCST)Gold Coast

Divisekera S (2009) lsquoEx-post demand for Australian tourism goods and servicesrsquo Tourism EconomicsVol 15 No 1 pp 153ndash180

Divisekera S (2010) lsquoEconomics of leisure and non-leisure tourist demand a study of domesticdemand for Australian tourismrsquo Tourism Economics Vol 16 No 1 pp 117ndash136

Dolnicar S Crouch GI Devinney T Huybers T Louviere JJ and Oppewal H (2008)lsquoTourism and discretionary income allocation Heterogeneity among householdsrsquo TourismManagement Vol 29 No 1 pp 44ndash52

Downward P and Lumsdon L (2000) lsquoThe demand for day-visits an analysis of visitor spendingrsquoTourism Economics Vol 6 No 3 pp 251ndash261

23Determinants in tourist expenditure composition

Table 3 Frequency distributions of endogenous variables (not included in the MANOVAmodel)

Variables Count Percentage

ActivitiesGolfYes 227 12No 19132 988HikingYes 364 19No 18995 981Sporting eventsYes 603 31No 18756 969Nautical sportsYes 1645 85No 17714 915Other sportsYes 904 47No 18455 953Cultural visitsYes 11180 578No 8179 422Cultural eventsYes 2452 127No 16907 873Other cultural activitiesYes 4175 216No 15184 784SpaYes 1199 62No 18160 938Theme parksYes 1745 90No 17614 910GastronomyYes 1565 81No 17794 919Type of accommodationHotel 4ndash5 5056 261Hotel 3 4922 254Hotel lt 3 6673 345Rented appartment 2116 109Other accommodation 592 31Length of stay5 or less nights 9911 5126ndash8 nights 5952 3079ndash12 nights 1592 8213ndash15 nights 1261 6516 and more nights 643 33

TOURISM ECONOMICS24

Table 3 continued

Variables Count Percentage

Total expenditureQuartile 1 5074 262Quartile 2 5104 264Quartile 3 4986 258Quartile 4 4195 217Expenditure at destination per dayQuartile 1 4855 251Quartile 2 4832 250Quartile 3 4868 251Quartile 4 4804 248Motivation-destinationLeisure on coastin countryside 8083 418Urban tourism 5324 275Others 5952 307

Level of income in contrast affects both log-ratios The higher the incomethe lower the transport share The higher the income the lower the basicdiscretionary expense ratio that is a higher portion of the non-transportexpense is devoted to the discretionary budget

As far as the country of residence is concerned the results show that it affectsmainly the basicdiscretionary log-ratio and exhibits strong airline typemoderating effects LCA users tend to differ depending on their country ofresidence when distributing non-transport expenses into basic versusdiscretionary In general when compared to the UK and Ireland residentcategory (reference category) all other European residents have a largerdiscretionary share More precisely the French Italians and Portuguese have thehighest discretionary share within LCA users and other European countries(mainly central and eastern European) have the highest discretionary sharewithin legacy airline users

Gender has a small effect which is constant for both types of airline Femaletravellers tend to slightly increase the transport share and the share of basicexpenses within the non-transport expenses

The travel group seems to affect both log-ratios and has a moderating effectwith airline type Those tourists who travel in a family group have a highershare in transport expenses when flying with legacy airlines In the second log-ratio travelling as a family with legacy airlines increases the basic expensesshare compared to the share in discretionary Travelling with friends increasesthe share in non-transport expenses for both airline types Besides those whotravel with friends spend more in the discretionary share compared to basicexpenses and those who fly with LCA even more so Finally travelling aloneincreases the basic expenses share compared to the share in discretionaryexpenses for both airline types

When professional status is considered the results show that it mostly affectsthe basicdiscretionary log-ratio and there are some relevant airline type

25Determinants in tourist expenditure composition

Table 4 MANOVA results

y1 log-ratio y2 log-ratiotransportation basicat-destination discretionary

LCA Legacy LCA Legacy

Intercept 0225 0274 2195 1881

Up to high school 0192 0078 0406 0437University 0 0 0 0

Mediumlow income 0073 0289 0037 0518Medium income 0 0 0 0Mediumhigh income ndash0227 ndash0058 ndash0269 ndash0006High income ndash0495 ndash0190 ndash0463 ndash0141

Residence in Portugal 0041 0069 ndash1136 ndash0580Residence in other European Countries 0326 ndash0012 ndash0772 ndash0820Residence in Belgium ndash0127 ndash0159 ndash0667 ndash0461Residence in Netherlands ndash0120 ndash0197 ndash0678 ndash0530Residence in Austria + Switzerland +Liechtenstein ndash0017 ndash0087 ndash0687 ndash0436

Residence in France ndash0112 ndash0111 ndash0870 ndash0553Residence in Scandinavian countries ndash0045 ndash0038 ndash0660 ndash0477Residence in Germany ndash0283 ndash0176 ndash0452 ndash0392Residence in Italy ndash0209 ndash0266 ndash0855 ndash0503Residence in UK or Ireland 0 0 0 0

Female 0094 0094 0093 0093Male 0 0 0 0

Travelling alone ndash0053 ndash0002 0081 0126Travelling in family 0085 0193 0014 0213Travelling with friends ndash0237 ndash0191 ndash0504 ndash0282Travelling with partner 0 0 0 0

Homemaker 0080 0160 0253 0235Unemployed 0030 0159 ndash0036 ndash0039Student ndash0005 ndash0104 ndash0331 ndash0624Low-level employee ndash0034 0194 ndash0194 0063Mid-level employee 0 0 0 0High-level employee ndash0008 ndash0064 ndash0001 ndash0139Self-employed 0029 ndash0086 0193 ndash0072

Over 45 yearsrsquo old and pensioner ndash0060 ndash0060 0633 0633Over 45 yearsrsquo old and not pensioner ndash0010 ndash0010 0142 014225ndash44 yearsrsquo old 0 0 0 015ndash24 yearsrsquo old 0002 0002 ndash0019 ndash0019

moderating effects Legacy users with a low professional status (homemakerunemployed or low-level employee) spend more on transport compared to thenon-transport expenses Conversely students using legacy airlines spend moreon non-transport expenses and these non-transport expenses are to a greater

TOURISM ECONOMICS26

Figure 2 Endogenous variables on the log-ratio space Mean log-ratio valueswithin each category of the endogenous variables

extent devoted to discretionary expenses compared to basic expensesHomemakers increase basic expenditure compared to discretionary with respectto the reference category (mid-level employee) for both airline types LCA userswho are self-employed also increase basic expenditure compared to discretionaryLow-level employees flying with an LCA increase the discretionary sharecompared to the basic and high-level employees flying with legacy airlines alsospend relatively more on the discretionary part

The results show that age affects the second log-ratio more than the firstand there is no moderating effect with the airline type Being a pensioner resultsin a higher share in basic expenses compared to discretionary expenses for bothairline types The remaining categories have hardly any differences from thereference 25ndash44 yearsrsquo old

Results for the endogenous variables

As introduced in the methodology section Figure 2 shows how the endogenousvariables behave in the log-ratio space The origin of the graph is representedby the mean log-ratios in Table 1 01506 for the transportother ratio and17657 for the basicdiscretionary ratio and according to this the graph showscategories with higher or lower ratios than the average

075

060

045

030

015

000

ndash015

ndash030

ndash045

ndash060

Ratio tra

nsport

ationa

t-destination e

xpenses

Ratio basicdiscretionary expenses

050 075 100 125 150 175 200 225 250

27Determinants in tourist expenditure composition

Regarding accommodation it seems that rented apartments and other typesof accommodation behave similarly and have the lowest budget share intransport In the case of hotels as the star rating increases expenditure on basiccompared to discretionary increases and expenditure on transport compared toat-destination decreases

As far as the length of stay is concerned longer stays increase the share ofnon-transport expenditure in basic expenses and reduce the share in transport

As regards destination and motivation tourists going to a seaside or country-side destination for leisure purposes behave quite differently from those goingto an urban destination for urban or cultural tourism The former group spendsmore on transport and less on the discretionary portion However those comingfor urban tourism show an increase in the share of non-transport expenses aswell as in the discretionary share compared to basic expenses

The total expenditure variable gives relevant information It is ordered alongthe first log-ratio axis from top to bottom meaning that those tourists havinga lower total expenditure spend relatively more on the transport share and thosewho have a higher total expenditure spend more on the other two parts of thebudget Regarding the expenditure made at destination per day it is ordereddiagonally from the upper-right corner to the bottom-left corner meaning thattourists with a higher at-destination expenditure per day spend less ontransportation and at the same time more on the discretionary part within at-destination expenses

Finally activities undertaken are located in the lower left quadrant of thegraph meaning that undertaking activities (discretionary expenses) decreasesthe share in transport expenses and although they continue to spend more onbasic expenses than on discretionary the discretionary share increases We havehighlighted some of them Those tourists attending sports and cultural eventsare those spending the most on discretionary compared to basic expensesHiking or some nautical sports could be free of charge thus tourists doing theseactivities have a basic to discretionary ratio close to the mean As for thetransportation share all activities tend to reduce this share about equally withthe exception of golf which leads to a much more substantial reduction

Conclusion and discussion

The main purpose of this article was to study the composition of touristexpenditure and its determinants that is the drivers of the share of touristexpenditure allocated to the different categories of a travel budget and withspecial emphasis on the distinction between legacy versus LCA travellers

The main differences between legacy and LCA users are that income somecountries of residence the travel group and some occupations have differenteffects depending on the type of airline Tourists with a medium or low incomewho travel with legacy airlines tend to spend a greater share of their travelbudget on transportation and basic expenses but those travelling with LCAstend to spend more at the destination and more specifically ondiscretionary expenditure LCA tourists residing in the lsquoother Europeancountriesrsquo spend more on transportation Families flying with legacy airlinesspend more on the transportation share compared to the at-destination share

TOURISM ECONOMICS28

and also spend more on basic expenses compared to discretionary ones Finallyin terms of occupational status legacy users with low-level employment spendmore on transportation whereas the opposite occurs with LCA travellers whospend more on the discretionary component Students and high-level employeesflying with legacy airlines spend more on the discretionary part of the tripbudget

As far as implications for management are concerned assuming that theobjectives of DMOs are to increase the allocation to the at-destinationcomponent of the trip budget (that is lowering the first log-ratio) some of therecommended actions would be to increase marketing efforts directed atmedium- and high-income earners especially at those travelling on LCAs toincrease marketing strategies in some of the European outbound markets suchas Germany (especially those using LCA) and Italy and the Netherlands(especially those flying with legacy airlines) DMOs should also focus on thosetourists travelling with friends (airline type is inconsequential) and studentsusing legacy companies As for the endogenous variables DMOs should directtheir efforts towards those tourists undertaking activities towards those stayinga little bit longer than a week and towards those with high total expenditureFurthermore if DMOs are also interested in capturing the groups who spendrelatively more on discretionary expenses (lowering the second log-ratio) theyshould focus their marketing efforts on those tourists who are highly educatedas well as those who have medium to high incomes and who use LCAsAdditionally they should focus on markets other than the UK and IrelandYoung tourists and tourists who travel with friends (where airline type isinconsequential) low-level employees and students using LCAs as well asstudents and high-level employees flying on legacy airlines should also betargeted In relation to the endogenous variables DMOs must take into accountthose tourists undertaking payable activities those on shorter stays or comingfor urban and cultural tourism or other purposes

The appeal of the CODA methodology for studying tourism budgets lies inthe fact that once the variables have been transformed the researcher can usestandard and well understood statistical models with unbounded error termdistributions while ensuring that the results will be compositionally coherentand that the standard statistical assumptions will hold Conversely results arenot compositionally coherent when we use the MANOVA model on raw sharewithout transforming it into log-ratios We have seen with our data that 84of individuals included in the sample have at least one limit of the 90 shareprediction intervals outside the [01] range

Contrary to the classic and highly restrictive alr and clr transformations theCODA methodology offers the potential to construct tailor-made log-ratios whichare intuitive to interpret and suit the research questions at hand A classificationtree of components is a clear and useful tool in this respect We encourageresearchers to use this approach in further research on tourism expenditure whenbudget share division is fundamental to the researchersrsquo questions

In this study we have encountered two limitations First contrary to studiesthat use macro data in micro data studies the coexistence of full-board halfboard bed and breakfast and accommodation-only tourists makes it impossibleto separate accommodation and food expenditure meaningfully Othersubdivisions (for example discretionary expenditure on activities non-grocery

29Determinants in tourist expenditure composition

shopping and moving around the destination) are technically feasible but canonly increase the percentages of zeros The CODA methodology involves somedegree of amalgamation of components Second even though there are someadvantages to using a database from an official statistics institution as in thiscase (large sample size and scope) the main disadvantage is that the set ofavailable variables cannot be controlled by the researcher A further issue to beaddressed is the time dimension As the EGATUR survey is conducted annuallyfurther research can be done to include a repeat cross-section analysis in orderto capture trends in the effects of predictors

References

Aitchison J (1986) The Statistical Analysis of Compositional Data Monographs on Statistics and AppliedProbability Chapman and Hall London (Reprinted in 2003 with additional material by TheBlackburn Press)

Alegre J Mateo S and Pou LL (2009) lsquoParticipation in tourism consumption and the intensityof participation an analysis of their socio-demographic and economic determinantsrsquo TourismEconomics Vol 15 No 3 pp 531ndash546

Bewley R (1982) lsquoThe generalized addilog demand system applied to Australian time series andcross section datarsquo Australian Economic Papers Vol 21 No 38 pp 177ndash192

Bewley R and Fiebig DG (1988) lsquoEstimation of price elasticities for an international telephonedemand modelrsquo The Journal of Industrial Economics Vol 36 No 4 pp 393ndash409

Cai LA (1998) lsquoAnalysing household food expenditure patterns on trips and vacations a Tobitmodersquo Journal of Hospitality and Tourism Research Vol 22 No 4 pp 338ndash358

Cai LA (1999) lsquoRelationship of household characteristics and lodging expenditure on leisure tripsrsquoJournal of Hospitality and Leisure Marketing Vol 6 No 2 pp 5ndash18

Cai LA Hong G-S and Morrison AM (1995) lsquoHousehold expenditure patterns for tourismproducts and servicesrsquo Journal of Travel and Tourism Marketing Vol 4 No 4 pp 15ndash39

Cannon TF and Ford J (2002) lsquoRelationship of demographic and trip characteristics to visitorspending an analysis of sports travel visitors across timersquo Tourism Economics Vol 8 No 3 pp 263ndash271

Castillo-Manzano JL and Marchena-Goacutemez M (2010) lsquoAnalysis of determinants of airline choiceprofiling the LCC passengerrsquo Applied Economics Letters Vol 18 No 1 pp 49ndash53

Chiou Y-C and Chen Y-H (2010) lsquoFactors influencing the intentions of passengers regarding full ser-vice and low cost carriers a notersquo Journal of Air Transport Management Vol 16 No 4 pp 226ndash228

Coenen M and van Eekeren L (2003) lsquoA study of the demand for domestic tourism by Swedishhouseholds using a two-staged budgeting modelrsquo Scandinavian Journal of Hospitality and TourismVol 3 No 2 pp 114ndash133

Craggs R and Schofield P (2009) lsquoExpenditure-based segmentation and visitor profiling at theQuays in Salford UKrsquo Tourism Economics Vol 15 No 1 pp 243ndash260

Deaton A and Muellbauer J (1980) lsquoAn almost ideal demand systemrsquo American Economic ReviewVol 70 No 3 pp 312ndash326

Dellaert B Borgers A and Timmermans H (1997) lsquoConsumer activity pattern choice devel-opment and test of stage-dependent conjoint choice experimentsrsquo Journal of Retailing and ConsumerServices Vol 4 No 1 pp 25ndash37

Divisekera S (2007) Modelling and Estimation of Tourism Demand Elasticities A Study of TourismExpenditure Allocation in Australia Cooperative Research Centre for Sustainable Tourism (CRCST)Gold Coast

Divisekera S (2009) lsquoEx-post demand for Australian tourism goods and servicesrsquo Tourism EconomicsVol 15 No 1 pp 153ndash180

Divisekera S (2010) lsquoEconomics of leisure and non-leisure tourist demand a study of domesticdemand for Australian tourismrsquo Tourism Economics Vol 16 No 1 pp 117ndash136

Dolnicar S Crouch GI Devinney T Huybers T Louviere JJ and Oppewal H (2008)lsquoTourism and discretionary income allocation Heterogeneity among householdsrsquo TourismManagement Vol 29 No 1 pp 44ndash52

Downward P and Lumsdon L (2000) lsquoThe demand for day-visits an analysis of visitor spendingrsquoTourism Economics Vol 6 No 3 pp 251ndash261

TOURISM ECONOMICS24

Table 3 continued

Variables Count Percentage

Total expenditureQuartile 1 5074 262Quartile 2 5104 264Quartile 3 4986 258Quartile 4 4195 217Expenditure at destination per dayQuartile 1 4855 251Quartile 2 4832 250Quartile 3 4868 251Quartile 4 4804 248Motivation-destinationLeisure on coastin countryside 8083 418Urban tourism 5324 275Others 5952 307

Level of income in contrast affects both log-ratios The higher the incomethe lower the transport share The higher the income the lower the basicdiscretionary expense ratio that is a higher portion of the non-transportexpense is devoted to the discretionary budget

As far as the country of residence is concerned the results show that it affectsmainly the basicdiscretionary log-ratio and exhibits strong airline typemoderating effects LCA users tend to differ depending on their country ofresidence when distributing non-transport expenses into basic versusdiscretionary In general when compared to the UK and Ireland residentcategory (reference category) all other European residents have a largerdiscretionary share More precisely the French Italians and Portuguese have thehighest discretionary share within LCA users and other European countries(mainly central and eastern European) have the highest discretionary sharewithin legacy airline users

Gender has a small effect which is constant for both types of airline Femaletravellers tend to slightly increase the transport share and the share of basicexpenses within the non-transport expenses

The travel group seems to affect both log-ratios and has a moderating effectwith airline type Those tourists who travel in a family group have a highershare in transport expenses when flying with legacy airlines In the second log-ratio travelling as a family with legacy airlines increases the basic expensesshare compared to the share in discretionary Travelling with friends increasesthe share in non-transport expenses for both airline types Besides those whotravel with friends spend more in the discretionary share compared to basicexpenses and those who fly with LCA even more so Finally travelling aloneincreases the basic expenses share compared to the share in discretionaryexpenses for both airline types

When professional status is considered the results show that it mostly affectsthe basicdiscretionary log-ratio and there are some relevant airline type

25Determinants in tourist expenditure composition

Table 4 MANOVA results

y1 log-ratio y2 log-ratiotransportation basicat-destination discretionary

LCA Legacy LCA Legacy

Intercept 0225 0274 2195 1881

Up to high school 0192 0078 0406 0437University 0 0 0 0

Mediumlow income 0073 0289 0037 0518Medium income 0 0 0 0Mediumhigh income ndash0227 ndash0058 ndash0269 ndash0006High income ndash0495 ndash0190 ndash0463 ndash0141

Residence in Portugal 0041 0069 ndash1136 ndash0580Residence in other European Countries 0326 ndash0012 ndash0772 ndash0820Residence in Belgium ndash0127 ndash0159 ndash0667 ndash0461Residence in Netherlands ndash0120 ndash0197 ndash0678 ndash0530Residence in Austria + Switzerland +Liechtenstein ndash0017 ndash0087 ndash0687 ndash0436

Residence in France ndash0112 ndash0111 ndash0870 ndash0553Residence in Scandinavian countries ndash0045 ndash0038 ndash0660 ndash0477Residence in Germany ndash0283 ndash0176 ndash0452 ndash0392Residence in Italy ndash0209 ndash0266 ndash0855 ndash0503Residence in UK or Ireland 0 0 0 0

Female 0094 0094 0093 0093Male 0 0 0 0

Travelling alone ndash0053 ndash0002 0081 0126Travelling in family 0085 0193 0014 0213Travelling with friends ndash0237 ndash0191 ndash0504 ndash0282Travelling with partner 0 0 0 0

Homemaker 0080 0160 0253 0235Unemployed 0030 0159 ndash0036 ndash0039Student ndash0005 ndash0104 ndash0331 ndash0624Low-level employee ndash0034 0194 ndash0194 0063Mid-level employee 0 0 0 0High-level employee ndash0008 ndash0064 ndash0001 ndash0139Self-employed 0029 ndash0086 0193 ndash0072

Over 45 yearsrsquo old and pensioner ndash0060 ndash0060 0633 0633Over 45 yearsrsquo old and not pensioner ndash0010 ndash0010 0142 014225ndash44 yearsrsquo old 0 0 0 015ndash24 yearsrsquo old 0002 0002 ndash0019 ndash0019

moderating effects Legacy users with a low professional status (homemakerunemployed or low-level employee) spend more on transport compared to thenon-transport expenses Conversely students using legacy airlines spend moreon non-transport expenses and these non-transport expenses are to a greater

TOURISM ECONOMICS26

Figure 2 Endogenous variables on the log-ratio space Mean log-ratio valueswithin each category of the endogenous variables

extent devoted to discretionary expenses compared to basic expensesHomemakers increase basic expenditure compared to discretionary with respectto the reference category (mid-level employee) for both airline types LCA userswho are self-employed also increase basic expenditure compared to discretionaryLow-level employees flying with an LCA increase the discretionary sharecompared to the basic and high-level employees flying with legacy airlines alsospend relatively more on the discretionary part

The results show that age affects the second log-ratio more than the firstand there is no moderating effect with the airline type Being a pensioner resultsin a higher share in basic expenses compared to discretionary expenses for bothairline types The remaining categories have hardly any differences from thereference 25ndash44 yearsrsquo old

Results for the endogenous variables

As introduced in the methodology section Figure 2 shows how the endogenousvariables behave in the log-ratio space The origin of the graph is representedby the mean log-ratios in Table 1 01506 for the transportother ratio and17657 for the basicdiscretionary ratio and according to this the graph showscategories with higher or lower ratios than the average

075

060

045

030

015

000

ndash015

ndash030

ndash045

ndash060

Ratio tra

nsport

ationa

t-destination e

xpenses

Ratio basicdiscretionary expenses

050 075 100 125 150 175 200 225 250

27Determinants in tourist expenditure composition

Regarding accommodation it seems that rented apartments and other typesof accommodation behave similarly and have the lowest budget share intransport In the case of hotels as the star rating increases expenditure on basiccompared to discretionary increases and expenditure on transport compared toat-destination decreases

As far as the length of stay is concerned longer stays increase the share ofnon-transport expenditure in basic expenses and reduce the share in transport

As regards destination and motivation tourists going to a seaside or country-side destination for leisure purposes behave quite differently from those goingto an urban destination for urban or cultural tourism The former group spendsmore on transport and less on the discretionary portion However those comingfor urban tourism show an increase in the share of non-transport expenses aswell as in the discretionary share compared to basic expenses

The total expenditure variable gives relevant information It is ordered alongthe first log-ratio axis from top to bottom meaning that those tourists havinga lower total expenditure spend relatively more on the transport share and thosewho have a higher total expenditure spend more on the other two parts of thebudget Regarding the expenditure made at destination per day it is ordereddiagonally from the upper-right corner to the bottom-left corner meaning thattourists with a higher at-destination expenditure per day spend less ontransportation and at the same time more on the discretionary part within at-destination expenses

Finally activities undertaken are located in the lower left quadrant of thegraph meaning that undertaking activities (discretionary expenses) decreasesthe share in transport expenses and although they continue to spend more onbasic expenses than on discretionary the discretionary share increases We havehighlighted some of them Those tourists attending sports and cultural eventsare those spending the most on discretionary compared to basic expensesHiking or some nautical sports could be free of charge thus tourists doing theseactivities have a basic to discretionary ratio close to the mean As for thetransportation share all activities tend to reduce this share about equally withthe exception of golf which leads to a much more substantial reduction

Conclusion and discussion

The main purpose of this article was to study the composition of touristexpenditure and its determinants that is the drivers of the share of touristexpenditure allocated to the different categories of a travel budget and withspecial emphasis on the distinction between legacy versus LCA travellers

The main differences between legacy and LCA users are that income somecountries of residence the travel group and some occupations have differenteffects depending on the type of airline Tourists with a medium or low incomewho travel with legacy airlines tend to spend a greater share of their travelbudget on transportation and basic expenses but those travelling with LCAstend to spend more at the destination and more specifically ondiscretionary expenditure LCA tourists residing in the lsquoother Europeancountriesrsquo spend more on transportation Families flying with legacy airlinesspend more on the transportation share compared to the at-destination share

TOURISM ECONOMICS28

and also spend more on basic expenses compared to discretionary ones Finallyin terms of occupational status legacy users with low-level employment spendmore on transportation whereas the opposite occurs with LCA travellers whospend more on the discretionary component Students and high-level employeesflying with legacy airlines spend more on the discretionary part of the tripbudget

As far as implications for management are concerned assuming that theobjectives of DMOs are to increase the allocation to the at-destinationcomponent of the trip budget (that is lowering the first log-ratio) some of therecommended actions would be to increase marketing efforts directed atmedium- and high-income earners especially at those travelling on LCAs toincrease marketing strategies in some of the European outbound markets suchas Germany (especially those using LCA) and Italy and the Netherlands(especially those flying with legacy airlines) DMOs should also focus on thosetourists travelling with friends (airline type is inconsequential) and studentsusing legacy companies As for the endogenous variables DMOs should directtheir efforts towards those tourists undertaking activities towards those stayinga little bit longer than a week and towards those with high total expenditureFurthermore if DMOs are also interested in capturing the groups who spendrelatively more on discretionary expenses (lowering the second log-ratio) theyshould focus their marketing efforts on those tourists who are highly educatedas well as those who have medium to high incomes and who use LCAsAdditionally they should focus on markets other than the UK and IrelandYoung tourists and tourists who travel with friends (where airline type isinconsequential) low-level employees and students using LCAs as well asstudents and high-level employees flying on legacy airlines should also betargeted In relation to the endogenous variables DMOs must take into accountthose tourists undertaking payable activities those on shorter stays or comingfor urban and cultural tourism or other purposes

The appeal of the CODA methodology for studying tourism budgets lies inthe fact that once the variables have been transformed the researcher can usestandard and well understood statistical models with unbounded error termdistributions while ensuring that the results will be compositionally coherentand that the standard statistical assumptions will hold Conversely results arenot compositionally coherent when we use the MANOVA model on raw sharewithout transforming it into log-ratios We have seen with our data that 84of individuals included in the sample have at least one limit of the 90 shareprediction intervals outside the [01] range

Contrary to the classic and highly restrictive alr and clr transformations theCODA methodology offers the potential to construct tailor-made log-ratios whichare intuitive to interpret and suit the research questions at hand A classificationtree of components is a clear and useful tool in this respect We encourageresearchers to use this approach in further research on tourism expenditure whenbudget share division is fundamental to the researchersrsquo questions

In this study we have encountered two limitations First contrary to studiesthat use macro data in micro data studies the coexistence of full-board halfboard bed and breakfast and accommodation-only tourists makes it impossibleto separate accommodation and food expenditure meaningfully Othersubdivisions (for example discretionary expenditure on activities non-grocery

29Determinants in tourist expenditure composition

shopping and moving around the destination) are technically feasible but canonly increase the percentages of zeros The CODA methodology involves somedegree of amalgamation of components Second even though there are someadvantages to using a database from an official statistics institution as in thiscase (large sample size and scope) the main disadvantage is that the set ofavailable variables cannot be controlled by the researcher A further issue to beaddressed is the time dimension As the EGATUR survey is conducted annuallyfurther research can be done to include a repeat cross-section analysis in orderto capture trends in the effects of predictors

References

Aitchison J (1986) The Statistical Analysis of Compositional Data Monographs on Statistics and AppliedProbability Chapman and Hall London (Reprinted in 2003 with additional material by TheBlackburn Press)

Alegre J Mateo S and Pou LL (2009) lsquoParticipation in tourism consumption and the intensityof participation an analysis of their socio-demographic and economic determinantsrsquo TourismEconomics Vol 15 No 3 pp 531ndash546

Bewley R (1982) lsquoThe generalized addilog demand system applied to Australian time series andcross section datarsquo Australian Economic Papers Vol 21 No 38 pp 177ndash192

Bewley R and Fiebig DG (1988) lsquoEstimation of price elasticities for an international telephonedemand modelrsquo The Journal of Industrial Economics Vol 36 No 4 pp 393ndash409

Cai LA (1998) lsquoAnalysing household food expenditure patterns on trips and vacations a Tobitmodersquo Journal of Hospitality and Tourism Research Vol 22 No 4 pp 338ndash358

Cai LA (1999) lsquoRelationship of household characteristics and lodging expenditure on leisure tripsrsquoJournal of Hospitality and Leisure Marketing Vol 6 No 2 pp 5ndash18

Cai LA Hong G-S and Morrison AM (1995) lsquoHousehold expenditure patterns for tourismproducts and servicesrsquo Journal of Travel and Tourism Marketing Vol 4 No 4 pp 15ndash39

Cannon TF and Ford J (2002) lsquoRelationship of demographic and trip characteristics to visitorspending an analysis of sports travel visitors across timersquo Tourism Economics Vol 8 No 3 pp 263ndash271

Castillo-Manzano JL and Marchena-Goacutemez M (2010) lsquoAnalysis of determinants of airline choiceprofiling the LCC passengerrsquo Applied Economics Letters Vol 18 No 1 pp 49ndash53

Chiou Y-C and Chen Y-H (2010) lsquoFactors influencing the intentions of passengers regarding full ser-vice and low cost carriers a notersquo Journal of Air Transport Management Vol 16 No 4 pp 226ndash228

Coenen M and van Eekeren L (2003) lsquoA study of the demand for domestic tourism by Swedishhouseholds using a two-staged budgeting modelrsquo Scandinavian Journal of Hospitality and TourismVol 3 No 2 pp 114ndash133

Craggs R and Schofield P (2009) lsquoExpenditure-based segmentation and visitor profiling at theQuays in Salford UKrsquo Tourism Economics Vol 15 No 1 pp 243ndash260

Deaton A and Muellbauer J (1980) lsquoAn almost ideal demand systemrsquo American Economic ReviewVol 70 No 3 pp 312ndash326

Dellaert B Borgers A and Timmermans H (1997) lsquoConsumer activity pattern choice devel-opment and test of stage-dependent conjoint choice experimentsrsquo Journal of Retailing and ConsumerServices Vol 4 No 1 pp 25ndash37

Divisekera S (2007) Modelling and Estimation of Tourism Demand Elasticities A Study of TourismExpenditure Allocation in Australia Cooperative Research Centre for Sustainable Tourism (CRCST)Gold Coast

Divisekera S (2009) lsquoEx-post demand for Australian tourism goods and servicesrsquo Tourism EconomicsVol 15 No 1 pp 153ndash180

Divisekera S (2010) lsquoEconomics of leisure and non-leisure tourist demand a study of domesticdemand for Australian tourismrsquo Tourism Economics Vol 16 No 1 pp 117ndash136

Dolnicar S Crouch GI Devinney T Huybers T Louviere JJ and Oppewal H (2008)lsquoTourism and discretionary income allocation Heterogeneity among householdsrsquo TourismManagement Vol 29 No 1 pp 44ndash52

Downward P and Lumsdon L (2000) lsquoThe demand for day-visits an analysis of visitor spendingrsquoTourism Economics Vol 6 No 3 pp 251ndash261

25Determinants in tourist expenditure composition

Table 4 MANOVA results

y1 log-ratio y2 log-ratiotransportation basicat-destination discretionary

LCA Legacy LCA Legacy

Intercept 0225 0274 2195 1881

Up to high school 0192 0078 0406 0437University 0 0 0 0

Mediumlow income 0073 0289 0037 0518Medium income 0 0 0 0Mediumhigh income ndash0227 ndash0058 ndash0269 ndash0006High income ndash0495 ndash0190 ndash0463 ndash0141

Residence in Portugal 0041 0069 ndash1136 ndash0580Residence in other European Countries 0326 ndash0012 ndash0772 ndash0820Residence in Belgium ndash0127 ndash0159 ndash0667 ndash0461Residence in Netherlands ndash0120 ndash0197 ndash0678 ndash0530Residence in Austria + Switzerland +Liechtenstein ndash0017 ndash0087 ndash0687 ndash0436

Residence in France ndash0112 ndash0111 ndash0870 ndash0553Residence in Scandinavian countries ndash0045 ndash0038 ndash0660 ndash0477Residence in Germany ndash0283 ndash0176 ndash0452 ndash0392Residence in Italy ndash0209 ndash0266 ndash0855 ndash0503Residence in UK or Ireland 0 0 0 0

Female 0094 0094 0093 0093Male 0 0 0 0

Travelling alone ndash0053 ndash0002 0081 0126Travelling in family 0085 0193 0014 0213Travelling with friends ndash0237 ndash0191 ndash0504 ndash0282Travelling with partner 0 0 0 0

Homemaker 0080 0160 0253 0235Unemployed 0030 0159 ndash0036 ndash0039Student ndash0005 ndash0104 ndash0331 ndash0624Low-level employee ndash0034 0194 ndash0194 0063Mid-level employee 0 0 0 0High-level employee ndash0008 ndash0064 ndash0001 ndash0139Self-employed 0029 ndash0086 0193 ndash0072

Over 45 yearsrsquo old and pensioner ndash0060 ndash0060 0633 0633Over 45 yearsrsquo old and not pensioner ndash0010 ndash0010 0142 014225ndash44 yearsrsquo old 0 0 0 015ndash24 yearsrsquo old 0002 0002 ndash0019 ndash0019

moderating effects Legacy users with a low professional status (homemakerunemployed or low-level employee) spend more on transport compared to thenon-transport expenses Conversely students using legacy airlines spend moreon non-transport expenses and these non-transport expenses are to a greater

TOURISM ECONOMICS26

Figure 2 Endogenous variables on the log-ratio space Mean log-ratio valueswithin each category of the endogenous variables

extent devoted to discretionary expenses compared to basic expensesHomemakers increase basic expenditure compared to discretionary with respectto the reference category (mid-level employee) for both airline types LCA userswho are self-employed also increase basic expenditure compared to discretionaryLow-level employees flying with an LCA increase the discretionary sharecompared to the basic and high-level employees flying with legacy airlines alsospend relatively more on the discretionary part

The results show that age affects the second log-ratio more than the firstand there is no moderating effect with the airline type Being a pensioner resultsin a higher share in basic expenses compared to discretionary expenses for bothairline types The remaining categories have hardly any differences from thereference 25ndash44 yearsrsquo old

Results for the endogenous variables

As introduced in the methodology section Figure 2 shows how the endogenousvariables behave in the log-ratio space The origin of the graph is representedby the mean log-ratios in Table 1 01506 for the transportother ratio and17657 for the basicdiscretionary ratio and according to this the graph showscategories with higher or lower ratios than the average

075

060

045

030

015

000

ndash015

ndash030

ndash045

ndash060

Ratio tra

nsport

ationa

t-destination e

xpenses

Ratio basicdiscretionary expenses

050 075 100 125 150 175 200 225 250

27Determinants in tourist expenditure composition

Regarding accommodation it seems that rented apartments and other typesof accommodation behave similarly and have the lowest budget share intransport In the case of hotels as the star rating increases expenditure on basiccompared to discretionary increases and expenditure on transport compared toat-destination decreases

As far as the length of stay is concerned longer stays increase the share ofnon-transport expenditure in basic expenses and reduce the share in transport

As regards destination and motivation tourists going to a seaside or country-side destination for leisure purposes behave quite differently from those goingto an urban destination for urban or cultural tourism The former group spendsmore on transport and less on the discretionary portion However those comingfor urban tourism show an increase in the share of non-transport expenses aswell as in the discretionary share compared to basic expenses

The total expenditure variable gives relevant information It is ordered alongthe first log-ratio axis from top to bottom meaning that those tourists havinga lower total expenditure spend relatively more on the transport share and thosewho have a higher total expenditure spend more on the other two parts of thebudget Regarding the expenditure made at destination per day it is ordereddiagonally from the upper-right corner to the bottom-left corner meaning thattourists with a higher at-destination expenditure per day spend less ontransportation and at the same time more on the discretionary part within at-destination expenses

Finally activities undertaken are located in the lower left quadrant of thegraph meaning that undertaking activities (discretionary expenses) decreasesthe share in transport expenses and although they continue to spend more onbasic expenses than on discretionary the discretionary share increases We havehighlighted some of them Those tourists attending sports and cultural eventsare those spending the most on discretionary compared to basic expensesHiking or some nautical sports could be free of charge thus tourists doing theseactivities have a basic to discretionary ratio close to the mean As for thetransportation share all activities tend to reduce this share about equally withthe exception of golf which leads to a much more substantial reduction

Conclusion and discussion

The main purpose of this article was to study the composition of touristexpenditure and its determinants that is the drivers of the share of touristexpenditure allocated to the different categories of a travel budget and withspecial emphasis on the distinction between legacy versus LCA travellers

The main differences between legacy and LCA users are that income somecountries of residence the travel group and some occupations have differenteffects depending on the type of airline Tourists with a medium or low incomewho travel with legacy airlines tend to spend a greater share of their travelbudget on transportation and basic expenses but those travelling with LCAstend to spend more at the destination and more specifically ondiscretionary expenditure LCA tourists residing in the lsquoother Europeancountriesrsquo spend more on transportation Families flying with legacy airlinesspend more on the transportation share compared to the at-destination share

TOURISM ECONOMICS28

and also spend more on basic expenses compared to discretionary ones Finallyin terms of occupational status legacy users with low-level employment spendmore on transportation whereas the opposite occurs with LCA travellers whospend more on the discretionary component Students and high-level employeesflying with legacy airlines spend more on the discretionary part of the tripbudget

As far as implications for management are concerned assuming that theobjectives of DMOs are to increase the allocation to the at-destinationcomponent of the trip budget (that is lowering the first log-ratio) some of therecommended actions would be to increase marketing efforts directed atmedium- and high-income earners especially at those travelling on LCAs toincrease marketing strategies in some of the European outbound markets suchas Germany (especially those using LCA) and Italy and the Netherlands(especially those flying with legacy airlines) DMOs should also focus on thosetourists travelling with friends (airline type is inconsequential) and studentsusing legacy companies As for the endogenous variables DMOs should directtheir efforts towards those tourists undertaking activities towards those stayinga little bit longer than a week and towards those with high total expenditureFurthermore if DMOs are also interested in capturing the groups who spendrelatively more on discretionary expenses (lowering the second log-ratio) theyshould focus their marketing efforts on those tourists who are highly educatedas well as those who have medium to high incomes and who use LCAsAdditionally they should focus on markets other than the UK and IrelandYoung tourists and tourists who travel with friends (where airline type isinconsequential) low-level employees and students using LCAs as well asstudents and high-level employees flying on legacy airlines should also betargeted In relation to the endogenous variables DMOs must take into accountthose tourists undertaking payable activities those on shorter stays or comingfor urban and cultural tourism or other purposes

The appeal of the CODA methodology for studying tourism budgets lies inthe fact that once the variables have been transformed the researcher can usestandard and well understood statistical models with unbounded error termdistributions while ensuring that the results will be compositionally coherentand that the standard statistical assumptions will hold Conversely results arenot compositionally coherent when we use the MANOVA model on raw sharewithout transforming it into log-ratios We have seen with our data that 84of individuals included in the sample have at least one limit of the 90 shareprediction intervals outside the [01] range

Contrary to the classic and highly restrictive alr and clr transformations theCODA methodology offers the potential to construct tailor-made log-ratios whichare intuitive to interpret and suit the research questions at hand A classificationtree of components is a clear and useful tool in this respect We encourageresearchers to use this approach in further research on tourism expenditure whenbudget share division is fundamental to the researchersrsquo questions

In this study we have encountered two limitations First contrary to studiesthat use macro data in micro data studies the coexistence of full-board halfboard bed and breakfast and accommodation-only tourists makes it impossibleto separate accommodation and food expenditure meaningfully Othersubdivisions (for example discretionary expenditure on activities non-grocery

29Determinants in tourist expenditure composition

shopping and moving around the destination) are technically feasible but canonly increase the percentages of zeros The CODA methodology involves somedegree of amalgamation of components Second even though there are someadvantages to using a database from an official statistics institution as in thiscase (large sample size and scope) the main disadvantage is that the set ofavailable variables cannot be controlled by the researcher A further issue to beaddressed is the time dimension As the EGATUR survey is conducted annuallyfurther research can be done to include a repeat cross-section analysis in orderto capture trends in the effects of predictors

References

Aitchison J (1986) The Statistical Analysis of Compositional Data Monographs on Statistics and AppliedProbability Chapman and Hall London (Reprinted in 2003 with additional material by TheBlackburn Press)

Alegre J Mateo S and Pou LL (2009) lsquoParticipation in tourism consumption and the intensityof participation an analysis of their socio-demographic and economic determinantsrsquo TourismEconomics Vol 15 No 3 pp 531ndash546

Bewley R (1982) lsquoThe generalized addilog demand system applied to Australian time series andcross section datarsquo Australian Economic Papers Vol 21 No 38 pp 177ndash192

Bewley R and Fiebig DG (1988) lsquoEstimation of price elasticities for an international telephonedemand modelrsquo The Journal of Industrial Economics Vol 36 No 4 pp 393ndash409

Cai LA (1998) lsquoAnalysing household food expenditure patterns on trips and vacations a Tobitmodersquo Journal of Hospitality and Tourism Research Vol 22 No 4 pp 338ndash358

Cai LA (1999) lsquoRelationship of household characteristics and lodging expenditure on leisure tripsrsquoJournal of Hospitality and Leisure Marketing Vol 6 No 2 pp 5ndash18

Cai LA Hong G-S and Morrison AM (1995) lsquoHousehold expenditure patterns for tourismproducts and servicesrsquo Journal of Travel and Tourism Marketing Vol 4 No 4 pp 15ndash39

Cannon TF and Ford J (2002) lsquoRelationship of demographic and trip characteristics to visitorspending an analysis of sports travel visitors across timersquo Tourism Economics Vol 8 No 3 pp 263ndash271

Castillo-Manzano JL and Marchena-Goacutemez M (2010) lsquoAnalysis of determinants of airline choiceprofiling the LCC passengerrsquo Applied Economics Letters Vol 18 No 1 pp 49ndash53

Chiou Y-C and Chen Y-H (2010) lsquoFactors influencing the intentions of passengers regarding full ser-vice and low cost carriers a notersquo Journal of Air Transport Management Vol 16 No 4 pp 226ndash228

Coenen M and van Eekeren L (2003) lsquoA study of the demand for domestic tourism by Swedishhouseholds using a two-staged budgeting modelrsquo Scandinavian Journal of Hospitality and TourismVol 3 No 2 pp 114ndash133

Craggs R and Schofield P (2009) lsquoExpenditure-based segmentation and visitor profiling at theQuays in Salford UKrsquo Tourism Economics Vol 15 No 1 pp 243ndash260

Deaton A and Muellbauer J (1980) lsquoAn almost ideal demand systemrsquo American Economic ReviewVol 70 No 3 pp 312ndash326

Dellaert B Borgers A and Timmermans H (1997) lsquoConsumer activity pattern choice devel-opment and test of stage-dependent conjoint choice experimentsrsquo Journal of Retailing and ConsumerServices Vol 4 No 1 pp 25ndash37

Divisekera S (2007) Modelling and Estimation of Tourism Demand Elasticities A Study of TourismExpenditure Allocation in Australia Cooperative Research Centre for Sustainable Tourism (CRCST)Gold Coast

Divisekera S (2009) lsquoEx-post demand for Australian tourism goods and servicesrsquo Tourism EconomicsVol 15 No 1 pp 153ndash180

Divisekera S (2010) lsquoEconomics of leisure and non-leisure tourist demand a study of domesticdemand for Australian tourismrsquo Tourism Economics Vol 16 No 1 pp 117ndash136

Dolnicar S Crouch GI Devinney T Huybers T Louviere JJ and Oppewal H (2008)lsquoTourism and discretionary income allocation Heterogeneity among householdsrsquo TourismManagement Vol 29 No 1 pp 44ndash52

Downward P and Lumsdon L (2000) lsquoThe demand for day-visits an analysis of visitor spendingrsquoTourism Economics Vol 6 No 3 pp 251ndash261

TOURISM ECONOMICS26

Figure 2 Endogenous variables on the log-ratio space Mean log-ratio valueswithin each category of the endogenous variables

extent devoted to discretionary expenses compared to basic expensesHomemakers increase basic expenditure compared to discretionary with respectto the reference category (mid-level employee) for both airline types LCA userswho are self-employed also increase basic expenditure compared to discretionaryLow-level employees flying with an LCA increase the discretionary sharecompared to the basic and high-level employees flying with legacy airlines alsospend relatively more on the discretionary part

The results show that age affects the second log-ratio more than the firstand there is no moderating effect with the airline type Being a pensioner resultsin a higher share in basic expenses compared to discretionary expenses for bothairline types The remaining categories have hardly any differences from thereference 25ndash44 yearsrsquo old

Results for the endogenous variables

As introduced in the methodology section Figure 2 shows how the endogenousvariables behave in the log-ratio space The origin of the graph is representedby the mean log-ratios in Table 1 01506 for the transportother ratio and17657 for the basicdiscretionary ratio and according to this the graph showscategories with higher or lower ratios than the average

075

060

045

030

015

000

ndash015

ndash030

ndash045

ndash060

Ratio tra

nsport

ationa

t-destination e

xpenses

Ratio basicdiscretionary expenses

050 075 100 125 150 175 200 225 250

27Determinants in tourist expenditure composition

Regarding accommodation it seems that rented apartments and other typesof accommodation behave similarly and have the lowest budget share intransport In the case of hotels as the star rating increases expenditure on basiccompared to discretionary increases and expenditure on transport compared toat-destination decreases

As far as the length of stay is concerned longer stays increase the share ofnon-transport expenditure in basic expenses and reduce the share in transport

As regards destination and motivation tourists going to a seaside or country-side destination for leisure purposes behave quite differently from those goingto an urban destination for urban or cultural tourism The former group spendsmore on transport and less on the discretionary portion However those comingfor urban tourism show an increase in the share of non-transport expenses aswell as in the discretionary share compared to basic expenses

The total expenditure variable gives relevant information It is ordered alongthe first log-ratio axis from top to bottom meaning that those tourists havinga lower total expenditure spend relatively more on the transport share and thosewho have a higher total expenditure spend more on the other two parts of thebudget Regarding the expenditure made at destination per day it is ordereddiagonally from the upper-right corner to the bottom-left corner meaning thattourists with a higher at-destination expenditure per day spend less ontransportation and at the same time more on the discretionary part within at-destination expenses

Finally activities undertaken are located in the lower left quadrant of thegraph meaning that undertaking activities (discretionary expenses) decreasesthe share in transport expenses and although they continue to spend more onbasic expenses than on discretionary the discretionary share increases We havehighlighted some of them Those tourists attending sports and cultural eventsare those spending the most on discretionary compared to basic expensesHiking or some nautical sports could be free of charge thus tourists doing theseactivities have a basic to discretionary ratio close to the mean As for thetransportation share all activities tend to reduce this share about equally withthe exception of golf which leads to a much more substantial reduction

Conclusion and discussion

The main purpose of this article was to study the composition of touristexpenditure and its determinants that is the drivers of the share of touristexpenditure allocated to the different categories of a travel budget and withspecial emphasis on the distinction between legacy versus LCA travellers

The main differences between legacy and LCA users are that income somecountries of residence the travel group and some occupations have differenteffects depending on the type of airline Tourists with a medium or low incomewho travel with legacy airlines tend to spend a greater share of their travelbudget on transportation and basic expenses but those travelling with LCAstend to spend more at the destination and more specifically ondiscretionary expenditure LCA tourists residing in the lsquoother Europeancountriesrsquo spend more on transportation Families flying with legacy airlinesspend more on the transportation share compared to the at-destination share

TOURISM ECONOMICS28

and also spend more on basic expenses compared to discretionary ones Finallyin terms of occupational status legacy users with low-level employment spendmore on transportation whereas the opposite occurs with LCA travellers whospend more on the discretionary component Students and high-level employeesflying with legacy airlines spend more on the discretionary part of the tripbudget

As far as implications for management are concerned assuming that theobjectives of DMOs are to increase the allocation to the at-destinationcomponent of the trip budget (that is lowering the first log-ratio) some of therecommended actions would be to increase marketing efforts directed atmedium- and high-income earners especially at those travelling on LCAs toincrease marketing strategies in some of the European outbound markets suchas Germany (especially those using LCA) and Italy and the Netherlands(especially those flying with legacy airlines) DMOs should also focus on thosetourists travelling with friends (airline type is inconsequential) and studentsusing legacy companies As for the endogenous variables DMOs should directtheir efforts towards those tourists undertaking activities towards those stayinga little bit longer than a week and towards those with high total expenditureFurthermore if DMOs are also interested in capturing the groups who spendrelatively more on discretionary expenses (lowering the second log-ratio) theyshould focus their marketing efforts on those tourists who are highly educatedas well as those who have medium to high incomes and who use LCAsAdditionally they should focus on markets other than the UK and IrelandYoung tourists and tourists who travel with friends (where airline type isinconsequential) low-level employees and students using LCAs as well asstudents and high-level employees flying on legacy airlines should also betargeted In relation to the endogenous variables DMOs must take into accountthose tourists undertaking payable activities those on shorter stays or comingfor urban and cultural tourism or other purposes

The appeal of the CODA methodology for studying tourism budgets lies inthe fact that once the variables have been transformed the researcher can usestandard and well understood statistical models with unbounded error termdistributions while ensuring that the results will be compositionally coherentand that the standard statistical assumptions will hold Conversely results arenot compositionally coherent when we use the MANOVA model on raw sharewithout transforming it into log-ratios We have seen with our data that 84of individuals included in the sample have at least one limit of the 90 shareprediction intervals outside the [01] range

Contrary to the classic and highly restrictive alr and clr transformations theCODA methodology offers the potential to construct tailor-made log-ratios whichare intuitive to interpret and suit the research questions at hand A classificationtree of components is a clear and useful tool in this respect We encourageresearchers to use this approach in further research on tourism expenditure whenbudget share division is fundamental to the researchersrsquo questions

In this study we have encountered two limitations First contrary to studiesthat use macro data in micro data studies the coexistence of full-board halfboard bed and breakfast and accommodation-only tourists makes it impossibleto separate accommodation and food expenditure meaningfully Othersubdivisions (for example discretionary expenditure on activities non-grocery

29Determinants in tourist expenditure composition

shopping and moving around the destination) are technically feasible but canonly increase the percentages of zeros The CODA methodology involves somedegree of amalgamation of components Second even though there are someadvantages to using a database from an official statistics institution as in thiscase (large sample size and scope) the main disadvantage is that the set ofavailable variables cannot be controlled by the researcher A further issue to beaddressed is the time dimension As the EGATUR survey is conducted annuallyfurther research can be done to include a repeat cross-section analysis in orderto capture trends in the effects of predictors

References

Aitchison J (1986) The Statistical Analysis of Compositional Data Monographs on Statistics and AppliedProbability Chapman and Hall London (Reprinted in 2003 with additional material by TheBlackburn Press)

Alegre J Mateo S and Pou LL (2009) lsquoParticipation in tourism consumption and the intensityof participation an analysis of their socio-demographic and economic determinantsrsquo TourismEconomics Vol 15 No 3 pp 531ndash546

Bewley R (1982) lsquoThe generalized addilog demand system applied to Australian time series andcross section datarsquo Australian Economic Papers Vol 21 No 38 pp 177ndash192

Bewley R and Fiebig DG (1988) lsquoEstimation of price elasticities for an international telephonedemand modelrsquo The Journal of Industrial Economics Vol 36 No 4 pp 393ndash409

Cai LA (1998) lsquoAnalysing household food expenditure patterns on trips and vacations a Tobitmodersquo Journal of Hospitality and Tourism Research Vol 22 No 4 pp 338ndash358

Cai LA (1999) lsquoRelationship of household characteristics and lodging expenditure on leisure tripsrsquoJournal of Hospitality and Leisure Marketing Vol 6 No 2 pp 5ndash18

Cai LA Hong G-S and Morrison AM (1995) lsquoHousehold expenditure patterns for tourismproducts and servicesrsquo Journal of Travel and Tourism Marketing Vol 4 No 4 pp 15ndash39

Cannon TF and Ford J (2002) lsquoRelationship of demographic and trip characteristics to visitorspending an analysis of sports travel visitors across timersquo Tourism Economics Vol 8 No 3 pp 263ndash271

Castillo-Manzano JL and Marchena-Goacutemez M (2010) lsquoAnalysis of determinants of airline choiceprofiling the LCC passengerrsquo Applied Economics Letters Vol 18 No 1 pp 49ndash53

Chiou Y-C and Chen Y-H (2010) lsquoFactors influencing the intentions of passengers regarding full ser-vice and low cost carriers a notersquo Journal of Air Transport Management Vol 16 No 4 pp 226ndash228

Coenen M and van Eekeren L (2003) lsquoA study of the demand for domestic tourism by Swedishhouseholds using a two-staged budgeting modelrsquo Scandinavian Journal of Hospitality and TourismVol 3 No 2 pp 114ndash133

Craggs R and Schofield P (2009) lsquoExpenditure-based segmentation and visitor profiling at theQuays in Salford UKrsquo Tourism Economics Vol 15 No 1 pp 243ndash260

Deaton A and Muellbauer J (1980) lsquoAn almost ideal demand systemrsquo American Economic ReviewVol 70 No 3 pp 312ndash326

Dellaert B Borgers A and Timmermans H (1997) lsquoConsumer activity pattern choice devel-opment and test of stage-dependent conjoint choice experimentsrsquo Journal of Retailing and ConsumerServices Vol 4 No 1 pp 25ndash37

Divisekera S (2007) Modelling and Estimation of Tourism Demand Elasticities A Study of TourismExpenditure Allocation in Australia Cooperative Research Centre for Sustainable Tourism (CRCST)Gold Coast

Divisekera S (2009) lsquoEx-post demand for Australian tourism goods and servicesrsquo Tourism EconomicsVol 15 No 1 pp 153ndash180

Divisekera S (2010) lsquoEconomics of leisure and non-leisure tourist demand a study of domesticdemand for Australian tourismrsquo Tourism Economics Vol 16 No 1 pp 117ndash136

Dolnicar S Crouch GI Devinney T Huybers T Louviere JJ and Oppewal H (2008)lsquoTourism and discretionary income allocation Heterogeneity among householdsrsquo TourismManagement Vol 29 No 1 pp 44ndash52

Downward P and Lumsdon L (2000) lsquoThe demand for day-visits an analysis of visitor spendingrsquoTourism Economics Vol 6 No 3 pp 251ndash261

27Determinants in tourist expenditure composition

Regarding accommodation it seems that rented apartments and other typesof accommodation behave similarly and have the lowest budget share intransport In the case of hotels as the star rating increases expenditure on basiccompared to discretionary increases and expenditure on transport compared toat-destination decreases

As far as the length of stay is concerned longer stays increase the share ofnon-transport expenditure in basic expenses and reduce the share in transport

As regards destination and motivation tourists going to a seaside or country-side destination for leisure purposes behave quite differently from those goingto an urban destination for urban or cultural tourism The former group spendsmore on transport and less on the discretionary portion However those comingfor urban tourism show an increase in the share of non-transport expenses aswell as in the discretionary share compared to basic expenses

The total expenditure variable gives relevant information It is ordered alongthe first log-ratio axis from top to bottom meaning that those tourists havinga lower total expenditure spend relatively more on the transport share and thosewho have a higher total expenditure spend more on the other two parts of thebudget Regarding the expenditure made at destination per day it is ordereddiagonally from the upper-right corner to the bottom-left corner meaning thattourists with a higher at-destination expenditure per day spend less ontransportation and at the same time more on the discretionary part within at-destination expenses

Finally activities undertaken are located in the lower left quadrant of thegraph meaning that undertaking activities (discretionary expenses) decreasesthe share in transport expenses and although they continue to spend more onbasic expenses than on discretionary the discretionary share increases We havehighlighted some of them Those tourists attending sports and cultural eventsare those spending the most on discretionary compared to basic expensesHiking or some nautical sports could be free of charge thus tourists doing theseactivities have a basic to discretionary ratio close to the mean As for thetransportation share all activities tend to reduce this share about equally withthe exception of golf which leads to a much more substantial reduction

Conclusion and discussion

The main purpose of this article was to study the composition of touristexpenditure and its determinants that is the drivers of the share of touristexpenditure allocated to the different categories of a travel budget and withspecial emphasis on the distinction between legacy versus LCA travellers

The main differences between legacy and LCA users are that income somecountries of residence the travel group and some occupations have differenteffects depending on the type of airline Tourists with a medium or low incomewho travel with legacy airlines tend to spend a greater share of their travelbudget on transportation and basic expenses but those travelling with LCAstend to spend more at the destination and more specifically ondiscretionary expenditure LCA tourists residing in the lsquoother Europeancountriesrsquo spend more on transportation Families flying with legacy airlinesspend more on the transportation share compared to the at-destination share

TOURISM ECONOMICS28

and also spend more on basic expenses compared to discretionary ones Finallyin terms of occupational status legacy users with low-level employment spendmore on transportation whereas the opposite occurs with LCA travellers whospend more on the discretionary component Students and high-level employeesflying with legacy airlines spend more on the discretionary part of the tripbudget

As far as implications for management are concerned assuming that theobjectives of DMOs are to increase the allocation to the at-destinationcomponent of the trip budget (that is lowering the first log-ratio) some of therecommended actions would be to increase marketing efforts directed atmedium- and high-income earners especially at those travelling on LCAs toincrease marketing strategies in some of the European outbound markets suchas Germany (especially those using LCA) and Italy and the Netherlands(especially those flying with legacy airlines) DMOs should also focus on thosetourists travelling with friends (airline type is inconsequential) and studentsusing legacy companies As for the endogenous variables DMOs should directtheir efforts towards those tourists undertaking activities towards those stayinga little bit longer than a week and towards those with high total expenditureFurthermore if DMOs are also interested in capturing the groups who spendrelatively more on discretionary expenses (lowering the second log-ratio) theyshould focus their marketing efforts on those tourists who are highly educatedas well as those who have medium to high incomes and who use LCAsAdditionally they should focus on markets other than the UK and IrelandYoung tourists and tourists who travel with friends (where airline type isinconsequential) low-level employees and students using LCAs as well asstudents and high-level employees flying on legacy airlines should also betargeted In relation to the endogenous variables DMOs must take into accountthose tourists undertaking payable activities those on shorter stays or comingfor urban and cultural tourism or other purposes

The appeal of the CODA methodology for studying tourism budgets lies inthe fact that once the variables have been transformed the researcher can usestandard and well understood statistical models with unbounded error termdistributions while ensuring that the results will be compositionally coherentand that the standard statistical assumptions will hold Conversely results arenot compositionally coherent when we use the MANOVA model on raw sharewithout transforming it into log-ratios We have seen with our data that 84of individuals included in the sample have at least one limit of the 90 shareprediction intervals outside the [01] range

Contrary to the classic and highly restrictive alr and clr transformations theCODA methodology offers the potential to construct tailor-made log-ratios whichare intuitive to interpret and suit the research questions at hand A classificationtree of components is a clear and useful tool in this respect We encourageresearchers to use this approach in further research on tourism expenditure whenbudget share division is fundamental to the researchersrsquo questions

In this study we have encountered two limitations First contrary to studiesthat use macro data in micro data studies the coexistence of full-board halfboard bed and breakfast and accommodation-only tourists makes it impossibleto separate accommodation and food expenditure meaningfully Othersubdivisions (for example discretionary expenditure on activities non-grocery

29Determinants in tourist expenditure composition

shopping and moving around the destination) are technically feasible but canonly increase the percentages of zeros The CODA methodology involves somedegree of amalgamation of components Second even though there are someadvantages to using a database from an official statistics institution as in thiscase (large sample size and scope) the main disadvantage is that the set ofavailable variables cannot be controlled by the researcher A further issue to beaddressed is the time dimension As the EGATUR survey is conducted annuallyfurther research can be done to include a repeat cross-section analysis in orderto capture trends in the effects of predictors

References

Aitchison J (1986) The Statistical Analysis of Compositional Data Monographs on Statistics and AppliedProbability Chapman and Hall London (Reprinted in 2003 with additional material by TheBlackburn Press)

Alegre J Mateo S and Pou LL (2009) lsquoParticipation in tourism consumption and the intensityof participation an analysis of their socio-demographic and economic determinantsrsquo TourismEconomics Vol 15 No 3 pp 531ndash546

Bewley R (1982) lsquoThe generalized addilog demand system applied to Australian time series andcross section datarsquo Australian Economic Papers Vol 21 No 38 pp 177ndash192

Bewley R and Fiebig DG (1988) lsquoEstimation of price elasticities for an international telephonedemand modelrsquo The Journal of Industrial Economics Vol 36 No 4 pp 393ndash409

Cai LA (1998) lsquoAnalysing household food expenditure patterns on trips and vacations a Tobitmodersquo Journal of Hospitality and Tourism Research Vol 22 No 4 pp 338ndash358

Cai LA (1999) lsquoRelationship of household characteristics and lodging expenditure on leisure tripsrsquoJournal of Hospitality and Leisure Marketing Vol 6 No 2 pp 5ndash18

Cai LA Hong G-S and Morrison AM (1995) lsquoHousehold expenditure patterns for tourismproducts and servicesrsquo Journal of Travel and Tourism Marketing Vol 4 No 4 pp 15ndash39

Cannon TF and Ford J (2002) lsquoRelationship of demographic and trip characteristics to visitorspending an analysis of sports travel visitors across timersquo Tourism Economics Vol 8 No 3 pp 263ndash271

Castillo-Manzano JL and Marchena-Goacutemez M (2010) lsquoAnalysis of determinants of airline choiceprofiling the LCC passengerrsquo Applied Economics Letters Vol 18 No 1 pp 49ndash53

Chiou Y-C and Chen Y-H (2010) lsquoFactors influencing the intentions of passengers regarding full ser-vice and low cost carriers a notersquo Journal of Air Transport Management Vol 16 No 4 pp 226ndash228

Coenen M and van Eekeren L (2003) lsquoA study of the demand for domestic tourism by Swedishhouseholds using a two-staged budgeting modelrsquo Scandinavian Journal of Hospitality and TourismVol 3 No 2 pp 114ndash133

Craggs R and Schofield P (2009) lsquoExpenditure-based segmentation and visitor profiling at theQuays in Salford UKrsquo Tourism Economics Vol 15 No 1 pp 243ndash260

Deaton A and Muellbauer J (1980) lsquoAn almost ideal demand systemrsquo American Economic ReviewVol 70 No 3 pp 312ndash326

Dellaert B Borgers A and Timmermans H (1997) lsquoConsumer activity pattern choice devel-opment and test of stage-dependent conjoint choice experimentsrsquo Journal of Retailing and ConsumerServices Vol 4 No 1 pp 25ndash37

Divisekera S (2007) Modelling and Estimation of Tourism Demand Elasticities A Study of TourismExpenditure Allocation in Australia Cooperative Research Centre for Sustainable Tourism (CRCST)Gold Coast

Divisekera S (2009) lsquoEx-post demand for Australian tourism goods and servicesrsquo Tourism EconomicsVol 15 No 1 pp 153ndash180

Divisekera S (2010) lsquoEconomics of leisure and non-leisure tourist demand a study of domesticdemand for Australian tourismrsquo Tourism Economics Vol 16 No 1 pp 117ndash136

Dolnicar S Crouch GI Devinney T Huybers T Louviere JJ and Oppewal H (2008)lsquoTourism and discretionary income allocation Heterogeneity among householdsrsquo TourismManagement Vol 29 No 1 pp 44ndash52

Downward P and Lumsdon L (2000) lsquoThe demand for day-visits an analysis of visitor spendingrsquoTourism Economics Vol 6 No 3 pp 251ndash261

TOURISM ECONOMICS28

and also spend more on basic expenses compared to discretionary ones Finallyin terms of occupational status legacy users with low-level employment spendmore on transportation whereas the opposite occurs with LCA travellers whospend more on the discretionary component Students and high-level employeesflying with legacy airlines spend more on the discretionary part of the tripbudget

As far as implications for management are concerned assuming that theobjectives of DMOs are to increase the allocation to the at-destinationcomponent of the trip budget (that is lowering the first log-ratio) some of therecommended actions would be to increase marketing efforts directed atmedium- and high-income earners especially at those travelling on LCAs toincrease marketing strategies in some of the European outbound markets suchas Germany (especially those using LCA) and Italy and the Netherlands(especially those flying with legacy airlines) DMOs should also focus on thosetourists travelling with friends (airline type is inconsequential) and studentsusing legacy companies As for the endogenous variables DMOs should directtheir efforts towards those tourists undertaking activities towards those stayinga little bit longer than a week and towards those with high total expenditureFurthermore if DMOs are also interested in capturing the groups who spendrelatively more on discretionary expenses (lowering the second log-ratio) theyshould focus their marketing efforts on those tourists who are highly educatedas well as those who have medium to high incomes and who use LCAsAdditionally they should focus on markets other than the UK and IrelandYoung tourists and tourists who travel with friends (where airline type isinconsequential) low-level employees and students using LCAs as well asstudents and high-level employees flying on legacy airlines should also betargeted In relation to the endogenous variables DMOs must take into accountthose tourists undertaking payable activities those on shorter stays or comingfor urban and cultural tourism or other purposes

The appeal of the CODA methodology for studying tourism budgets lies inthe fact that once the variables have been transformed the researcher can usestandard and well understood statistical models with unbounded error termdistributions while ensuring that the results will be compositionally coherentand that the standard statistical assumptions will hold Conversely results arenot compositionally coherent when we use the MANOVA model on raw sharewithout transforming it into log-ratios We have seen with our data that 84of individuals included in the sample have at least one limit of the 90 shareprediction intervals outside the [01] range

Contrary to the classic and highly restrictive alr and clr transformations theCODA methodology offers the potential to construct tailor-made log-ratios whichare intuitive to interpret and suit the research questions at hand A classificationtree of components is a clear and useful tool in this respect We encourageresearchers to use this approach in further research on tourism expenditure whenbudget share division is fundamental to the researchersrsquo questions

In this study we have encountered two limitations First contrary to studiesthat use macro data in micro data studies the coexistence of full-board halfboard bed and breakfast and accommodation-only tourists makes it impossibleto separate accommodation and food expenditure meaningfully Othersubdivisions (for example discretionary expenditure on activities non-grocery

29Determinants in tourist expenditure composition

shopping and moving around the destination) are technically feasible but canonly increase the percentages of zeros The CODA methodology involves somedegree of amalgamation of components Second even though there are someadvantages to using a database from an official statistics institution as in thiscase (large sample size and scope) the main disadvantage is that the set ofavailable variables cannot be controlled by the researcher A further issue to beaddressed is the time dimension As the EGATUR survey is conducted annuallyfurther research can be done to include a repeat cross-section analysis in orderto capture trends in the effects of predictors

References

Aitchison J (1986) The Statistical Analysis of Compositional Data Monographs on Statistics and AppliedProbability Chapman and Hall London (Reprinted in 2003 with additional material by TheBlackburn Press)

Alegre J Mateo S and Pou LL (2009) lsquoParticipation in tourism consumption and the intensityof participation an analysis of their socio-demographic and economic determinantsrsquo TourismEconomics Vol 15 No 3 pp 531ndash546

Bewley R (1982) lsquoThe generalized addilog demand system applied to Australian time series andcross section datarsquo Australian Economic Papers Vol 21 No 38 pp 177ndash192

Bewley R and Fiebig DG (1988) lsquoEstimation of price elasticities for an international telephonedemand modelrsquo The Journal of Industrial Economics Vol 36 No 4 pp 393ndash409

Cai LA (1998) lsquoAnalysing household food expenditure patterns on trips and vacations a Tobitmodersquo Journal of Hospitality and Tourism Research Vol 22 No 4 pp 338ndash358

Cai LA (1999) lsquoRelationship of household characteristics and lodging expenditure on leisure tripsrsquoJournal of Hospitality and Leisure Marketing Vol 6 No 2 pp 5ndash18

Cai LA Hong G-S and Morrison AM (1995) lsquoHousehold expenditure patterns for tourismproducts and servicesrsquo Journal of Travel and Tourism Marketing Vol 4 No 4 pp 15ndash39

Cannon TF and Ford J (2002) lsquoRelationship of demographic and trip characteristics to visitorspending an analysis of sports travel visitors across timersquo Tourism Economics Vol 8 No 3 pp 263ndash271

Castillo-Manzano JL and Marchena-Goacutemez M (2010) lsquoAnalysis of determinants of airline choiceprofiling the LCC passengerrsquo Applied Economics Letters Vol 18 No 1 pp 49ndash53

Chiou Y-C and Chen Y-H (2010) lsquoFactors influencing the intentions of passengers regarding full ser-vice and low cost carriers a notersquo Journal of Air Transport Management Vol 16 No 4 pp 226ndash228

Coenen M and van Eekeren L (2003) lsquoA study of the demand for domestic tourism by Swedishhouseholds using a two-staged budgeting modelrsquo Scandinavian Journal of Hospitality and TourismVol 3 No 2 pp 114ndash133

Craggs R and Schofield P (2009) lsquoExpenditure-based segmentation and visitor profiling at theQuays in Salford UKrsquo Tourism Economics Vol 15 No 1 pp 243ndash260

Deaton A and Muellbauer J (1980) lsquoAn almost ideal demand systemrsquo American Economic ReviewVol 70 No 3 pp 312ndash326

Dellaert B Borgers A and Timmermans H (1997) lsquoConsumer activity pattern choice devel-opment and test of stage-dependent conjoint choice experimentsrsquo Journal of Retailing and ConsumerServices Vol 4 No 1 pp 25ndash37

Divisekera S (2007) Modelling and Estimation of Tourism Demand Elasticities A Study of TourismExpenditure Allocation in Australia Cooperative Research Centre for Sustainable Tourism (CRCST)Gold Coast

Divisekera S (2009) lsquoEx-post demand for Australian tourism goods and servicesrsquo Tourism EconomicsVol 15 No 1 pp 153ndash180

Divisekera S (2010) lsquoEconomics of leisure and non-leisure tourist demand a study of domesticdemand for Australian tourismrsquo Tourism Economics Vol 16 No 1 pp 117ndash136

Dolnicar S Crouch GI Devinney T Huybers T Louviere JJ and Oppewal H (2008)lsquoTourism and discretionary income allocation Heterogeneity among householdsrsquo TourismManagement Vol 29 No 1 pp 44ndash52

Downward P and Lumsdon L (2000) lsquoThe demand for day-visits an analysis of visitor spendingrsquoTourism Economics Vol 6 No 3 pp 251ndash261

29Determinants in tourist expenditure composition

shopping and moving around the destination) are technically feasible but canonly increase the percentages of zeros The CODA methodology involves somedegree of amalgamation of components Second even though there are someadvantages to using a database from an official statistics institution as in thiscase (large sample size and scope) the main disadvantage is that the set ofavailable variables cannot be controlled by the researcher A further issue to beaddressed is the time dimension As the EGATUR survey is conducted annuallyfurther research can be done to include a repeat cross-section analysis in orderto capture trends in the effects of predictors

References

Aitchison J (1986) The Statistical Analysis of Compositional Data Monographs on Statistics and AppliedProbability Chapman and Hall London (Reprinted in 2003 with additional material by TheBlackburn Press)

Alegre J Mateo S and Pou LL (2009) lsquoParticipation in tourism consumption and the intensityof participation an analysis of their socio-demographic and economic determinantsrsquo TourismEconomics Vol 15 No 3 pp 531ndash546

Bewley R (1982) lsquoThe generalized addilog demand system applied to Australian time series andcross section datarsquo Australian Economic Papers Vol 21 No 38 pp 177ndash192

Bewley R and Fiebig DG (1988) lsquoEstimation of price elasticities for an international telephonedemand modelrsquo The Journal of Industrial Economics Vol 36 No 4 pp 393ndash409

Cai LA (1998) lsquoAnalysing household food expenditure patterns on trips and vacations a Tobitmodersquo Journal of Hospitality and Tourism Research Vol 22 No 4 pp 338ndash358

Cai LA (1999) lsquoRelationship of household characteristics and lodging expenditure on leisure tripsrsquoJournal of Hospitality and Leisure Marketing Vol 6 No 2 pp 5ndash18

Cai LA Hong G-S and Morrison AM (1995) lsquoHousehold expenditure patterns for tourismproducts and servicesrsquo Journal of Travel and Tourism Marketing Vol 4 No 4 pp 15ndash39

Cannon TF and Ford J (2002) lsquoRelationship of demographic and trip characteristics to visitorspending an analysis of sports travel visitors across timersquo Tourism Economics Vol 8 No 3 pp 263ndash271

Castillo-Manzano JL and Marchena-Goacutemez M (2010) lsquoAnalysis of determinants of airline choiceprofiling the LCC passengerrsquo Applied Economics Letters Vol 18 No 1 pp 49ndash53

Chiou Y-C and Chen Y-H (2010) lsquoFactors influencing the intentions of passengers regarding full ser-vice and low cost carriers a notersquo Journal of Air Transport Management Vol 16 No 4 pp 226ndash228

Coenen M and van Eekeren L (2003) lsquoA study of the demand for domestic tourism by Swedishhouseholds using a two-staged budgeting modelrsquo Scandinavian Journal of Hospitality and TourismVol 3 No 2 pp 114ndash133

Craggs R and Schofield P (2009) lsquoExpenditure-based segmentation and visitor profiling at theQuays in Salford UKrsquo Tourism Economics Vol 15 No 1 pp 243ndash260

Deaton A and Muellbauer J (1980) lsquoAn almost ideal demand systemrsquo American Economic ReviewVol 70 No 3 pp 312ndash326

Dellaert B Borgers A and Timmermans H (1997) lsquoConsumer activity pattern choice devel-opment and test of stage-dependent conjoint choice experimentsrsquo Journal of Retailing and ConsumerServices Vol 4 No 1 pp 25ndash37

Divisekera S (2007) Modelling and Estimation of Tourism Demand Elasticities A Study of TourismExpenditure Allocation in Australia Cooperative Research Centre for Sustainable Tourism (CRCST)Gold Coast

Divisekera S (2009) lsquoEx-post demand for Australian tourism goods and servicesrsquo Tourism EconomicsVol 15 No 1 pp 153ndash180

Divisekera S (2010) lsquoEconomics of leisure and non-leisure tourist demand a study of domesticdemand for Australian tourismrsquo Tourism Economics Vol 16 No 1 pp 117ndash136

Dolnicar S Crouch GI Devinney T Huybers T Louviere JJ and Oppewal H (2008)lsquoTourism and discretionary income allocation Heterogeneity among householdsrsquo TourismManagement Vol 29 No 1 pp 44ndash52

Downward P and Lumsdon L (2000) lsquoThe demand for day-visits an analysis of visitor spendingrsquoTourism Economics Vol 6 No 3 pp 251ndash261

TOURISM ECONOMICS30

Downward P and Lumsdon L (2003) lsquoBeyond the demand for day-visits an analysis of visitorspendingrsquo Tourism Economics Vol 9 No 1 pp 67ndash76

Egozcue JJ and Pawlowsky-Glahn V (2005) lsquoGroups of parts and their balances in compositionaldata analysisrsquo Mathematical Geology Vol 37 No 7 pp 795ndash828

Egozcue JJ Pawlowsky-Glahn V Mateu-Figueras G and Barceloacute-Vidal C (2003) lsquoIsometriclogratio transformations for compositional data analysisrsquo Mathematical Geology Vol 35 No 3pp 279ndash300

Eugenio-Martin JL (2003) lsquoModelling determinants of tourism demand as a five-stage processa discrete choice methodological approachrsquo Tourism and Hospitality Research Vol 3 pp 341ndash354

Ferrer-Rosell B Martiacutenez-Garcia E and Coenders G (2014) lsquoPackage and no-frills air carriersas moderators of length of stayrsquo Tourism Management Vol 42 pp 114ndash122

Fesenmaier DR and Jeng J (2000) lsquoAssessing structure in the pleasure trip planning processrsquoTourism Analysis Vol 5 No 1 pp 13ndash27

Fleischer A Peleg G and Rivlin J (2011) lsquoThe impact of changes in household vacation expend-itures on the travel and hospitality industriesrsquo Tourism Management Vol 32 No 4 pp 815ndash821

Forgas S Moliner MA Saacutenchez J and Palau R (2010) lsquoAntecedents of airline passenger loyaltylow-cost versus traditional airlinesrsquo Journal of Air Transport Management Vol 16 No 4 pp 229ndash233

Fry T (2011) lsquoApplications in economicsrsquo in Pawlowsky-Glahn V and Buccianti A edsCompositional Data Analysis Theory and Applications Wiley New York pp 318ndash326

Fry JM Fry TRL and McLaren KR (1996) lsquoThe stochastic specification of demand shareequations restricting budget share to the unit simplexrsquo Journal of Econometrics Vol 73 No 2pp 377ndash385

Fry JM Fry TRL and McLaren KR (2000) lsquoCompositional data analysis and zeros in micro-datarsquo Applied Economics Vol 32 No 8 pp 953ndash959

Fujii ET Khaled M and Mak J (1985) lsquoAn almost ideal demand system for visitorexpendituresrsquo Journal of Transport Economics and Policy Vol 19 pp 161ndash171

Hong G-S Abdel-Ghany M and Kim SY (1999) lsquoModeling tourism spending decisions as atwo-step processrsquo European Advances in Consumer Research Vol 4 pp 216ndash223

Houthakker HS (1960) lsquoAdditive preferencesrsquo Econometrica Vol 28 No 2 pp 244ndash256Hung W-T Shang J-K and Wang F-C (2012) lsquoAnother look at the determinants of tourism

expenditurersquo Annals of Tourism Research Vol 39 No 1 pp 495ndash498IET (2004) Compantildeiacuteas Aeacutereas de Bajo Coste Informe Anual 2004 Instituto de Estudios Turiacutesticos

Ministerio de Industria Turismo y Comercio Madrid (httpwwwiettourspainesesndashESestadisticasotrasestadisticascompaniabajocosteanualesInforme20CompaC3B1C3ADas20Bajo20Coste202004pdf accessed 14 January 2013)

IET (2010) Compantildeiacuteas Aeacutereas de Bajo Coste Informe Anual 2010 Instituto de Estudios TuriacutesticosMinisterio de Industria Turismo y Comercio Madrid (httpwwwiettourspainesesndashESestadisticasotrasestadisticascompaniabajocosteanualesInforme20CompaC3B1C3ADas20Bajo20Coste202010pdf accessed 14 January 2013)

IET (2012) Egatur Encuesta de Gasto Turiacutestico Metodologia Instituto de Estudios Turiacutesticos Ministeriode Industria Energiacutea y Turismo Madrid (httpwwwiettourspainesesndashESestadisticasegaturmetodologiaReferencia20MetodolgicaNota20MetodolC3B3gica20Encuesta20de20Gasto20TurC3ADsticopdf accessed 22 November 2013)

Jang SCS Bai B Hong GS and OrsquoLeary JT (2004) lsquoUnderstanding travel expenditurepatterns a study of Japanese pleasure travellers to the United States by income levelrsquo TourismManagement Vol 25 No 3 pp 331ndash341

Kim YK and Lee HR (2011) lsquoCustomer satisfaction using low cost carriersrsquo Tourism ManagementVol 32 No 2 pp 235ndash243

Lee C (2001) lsquoDeterminants of recreational boater expenditures on tripsrsquo Tourism Management Vol22 No 6 pp 659ndash667

Legoheacuterel P (1998) lsquoToward a market segmentation of the tourism trade expenditure levels andconsumer behaviour instabilityrsquo Journal of Travel and Tourism Marketing Vol 7 No 3 pp 19ndash39

Lehto XY Morrison AM and OrsquoLeary JT (2001) lsquoDoes the visiting friends and relativesrsquotypology make a difference A study of the international VFR market to the United StatesrsquoJournal of Travel Research Vol 40 No 2 pp 201ndash212

Li G Song H and Witt SF (2004) lsquoModelling tourism demand a dynamic linear AIDSapproachrsquo Journal of Travel Research Vol 43 No 2 pp 141ndash150

31Determinants in tourist expenditure composition

Marcussen CH (2011) lsquoDeterminants of tourist spending in cross-sectional studies and at Danishdestinationsrsquo Tourism Economics Vol 17 No 4 pp 833ndash855

Martiacutenez-Garcia E and Raya JM (2008) lsquoLength of stay for low-cost tourismrsquo Tourism Manage-ment Vol 29 No 6 pp 1064ndash1075

Martiacutenez-Garcia E and Royo-Vela M (2010) lsquoSegmentation of low-cost flights users at secondaryairportsrsquo Journal of Air Transport Management Vol 16 pp 234ndash237

Martiacutenez-Garcia E Ferrer-Rosell B and Coenders G (2012) lsquoProfile of business and leisure travelerson low cost carriers in Europersquo Journal of Air Transport Management Vol 20 pp 12ndash14

Martiacuten-Fernaacutendez JA Barceloacute-Vidal C and Pawlowsky-Glahn V (2003) lsquoDealing with zeros andmissing values in compositional data sets using non-parametric imputationrsquo Mathematical GeologyVol 35 No 3 pp 253ndash278

Martiacuten-Fernaacutendez JA Palarea-Albaladejo J and Olea RA (2011) lsquoDealing with zerosrsquo inPawlowsky-Glahn V and Buccianti A eds Compositional Data Analysis Theory and ApplicationsWiley New York NY pp 47ndash62

Mateu-Figueras G Daunis-i-Estadella J and Martiacuten-Fernaacutendez JA (2013) lsquoAnalysing groupedCoDa Manova and post-hoc contrastsrsquo paper presented at the workshop on Compositional DataAnalysis (CoDaWork) Vorau 3ndash7 June

McLaren KR Fry JM and Fry TRL (1995) lsquoA simple nested test of the almost ideal demandsystemrsquo Empirical Economics Vol 20 No 1 pp 149ndash161

Melenberg B and Van Soest A (1996) lsquoParametric and semi-parametric modelling of Vacationexpendituresrsquo Journal of Applied Econometrics Vol 11 No 1 pp 59ndash76

Nicolau JL (2009) lsquoThe smile of the tourist the relationship between price sensitivity andexpensesrsquo The Service Industries Journal Vol 29 No 8 pp 1125ndash1134

Nicolau JL and Maacutes FJ (2005) lsquoHeckit modelling of tourist expenditure evidence from SpainrsquoInternational Journal of Service Industry Management Vol 16 No 3 pp 271ndash293

OrsquoConnell JF and Williams G (2005) lsquoPassengersrsquo perception of low cost airlines and full servicecarriers a case study involving Ryanair Aer Lingus Air Asia and Malaysia Airlinesrsquo Journal ofAir Transport Management Vol 11 pp 259ndash272

OrsquoHagan JW and Harrison MJ (1984) lsquoMarket share of US tourist expenditure in Europe aneconometric analysisrsquo Applied Economics Vol 16 pp 919ndash931

Oppermann M (1996) lsquoVisitation of tourism attractions and tourist expenditure patterns ndash repeatversus firstrsquo Asia Pacific Journal of Tourism Research Vol 1 No 1 pp 61ndash68

Pawlowsky-Glahn V and Buccianti A (2011) Compositional Data Analysis Theory and ApplicationsWiley New York NY

Pawlowsky-Glahn V and Egozcue JJ (2011) lsquoExploring compositional data with the CoDa-dendrogramrsquo Austrian Journal of Statistics Vol 40 No 1 pp 103ndash113

Pyo SS Uysal M and McLellan RW (1991) lsquoA linear expenditure model for tourism demandrsquoAnnals of Tourism Research Vol 18 No 3 pp 443ndash454

Raya-Vilchez JM and Martiacutenez-Garcia E (2011) lsquoNationality and low-cost trip duration Amicroeconometric analysisrsquo Journal of Air Transport Management Vol 17 pp 168ndash174

Ronning G (1992) lsquoShare equations in econometrics a story of repression frustration and deadendsrsquo Statistical Papers Vol 33 No 1 pp 307ndash334

Sainaghi R (2012) lsquoTourist expenditures the state of the artrsquo Anatolia An International Journalof Tourism and Hospitality Research Vol 23 No 2 pp 217ndash233

Song H Dwyer L and ZhengCao GL (2012) lsquoTourism economics research a review andassessmentrsquo Annals of Tourism Research Vol 39 No 3 pp 1653ndash1682

Syriopoulos TC and Sinclair T (1993) lsquoAn econometric study of tourism demand the AIDSmodel of US and European tourism in Mediterranean countriesrsquo Applied Economics Vol 25 No 12pp 1541ndash1552

Thioacute-Henestrosa S and Martiacuten-Fernaacutendez JA (2005) lsquoDealing with compositional data thefreeware CoDaPackrsquo Mathematical Geology Vol 37 No 7 pp 773ndash793

Wang Y and Davidson MCG (2010) lsquoA review of micro-analyses of tourist expenditurersquo CurrentIssues in Tourism Vol 13 No 6 pp 507ndash524

Wang Y Rompf P Severt D and Peerapatdit N (2006) lsquoExamining and identifying thedeterminants of travel expenditure patternsrsquo International Journal of Tourism Research Vol 8 No5 pp 333ndash346

Wu DC Li G and Song H (2011) lsquoAnalysing tourism consumption a dynamic system-of-equations approachrsquo Journal of Travel Research Vol 50 No 1 pp 46ndash56

TOURISM ECONOMICS32

Appendix

SPSS syntax code used to replace zeros

x1 x2 x3 contain raw share x1_prime x2_prime x3_prime contain zero-replaced share total_expenditure contains absolute expenditure delta is 1total_expenditure the zero replacement constant is k=065compute x1_prime = x1compute x2_prime = x2compute x3_prime = x3if (x3 = 0) x3_replacement = 0651total_expenditureif (x3 = 0) x3_prime = sum(x3 x3_replacement)if (X3 = 0) x2_prime = x2(1-x3_replacement)if (X3 = 0) x1_prime = x1(1-x3_replacement)execute