ARTICLE IN PRESS

0261-5177/$ - se

doi:10.1016/j.to

�CorrespondE-mail addr

(U. Mark).

Tourism Management 29 (2008) 469–486

www.elsevier.com/locate/tourman

Evaluating passive mobile positioning data for tourism surveys:An Estonian case study

Rein Ahasa,�, Anto Aasaa, Antti Roosea, Ular Markb, Siiri Silma

aInstitute of Geography, University of Tartu, Vanemuise 46, Tartu 51014, EstoniabPositium ICT, Niine 11, Tallinn 10414, Estonia

Received 9 June 2006; received in revised form 19 April 2007; accepted 16 May 2007

Abstract

This paper introduces the applicability of passive mobile positioning data in studying tourism. Passive mobile positioning data is

automatically stored in the memory files of mobile operators for call activities or movements of handsets in the network. For tourism

studies we use database of the locations of roaming (foreign phones) call activities in network cells: the location, time, random ID and

country of origin of the called phone. We describe the peculiarities of data, data gathering, sampling, the handling of the spatial database

and some analysis methods, using examples from Estonia. The results proved that mobile positioning data has valuable applications for

geographical studies. Correlations with conventional accommodation statistics in Estonia were up to 0.99 in the most commonly visited

tourist regions. Correlations of positioning data with accommodation statistics were lower in regions with a high number of transit

tourists and less tourism infrastructure. The results show that positioning data has advantages: data can be collected for larger spatial

units and in less visited areas; spatial and temporal preciseness is higher than for regular tourism statistics. Random IDs allow one to

study tourists’ movements, for example to study typical routes of tourists of certain nationalities. The weaknesses of data are related to

problems with accessing data, as operators do not wish to share data and because of privacy and surveillance concerns. Problem is also

that positioning data is another quantitative dataset with limited features.

r 2007 Elsevier Ltd. All rights reserved.

Keywords: Mobile positioning; Tourism; Geography; Space–time behaviour; Social positioning method; Estonia; Surveillance

1. Introduction

Tourism is of growing importance for the globaleconomy, since mobility has risen rapidly during recentdecades. Every tourism activity has a specific geographyand temporal sequence, and contemporary tourism geo-graphy has an increasing field of study in this matter. Inoperationalising tourism flows, most of the researchemphasises the supply–demand balance and visitors’behaviour, as well as their impacts on the physicalenvironment in recent decades (Hall & Page, 2006). Forthe recording of tourism flows there are different datasources and methods such as border, transportation andaccommodation statistics, questionnaire surveys and a

e front matter r 2007 Elsevier Ltd. All rights reserved.

urman.2007.05.014

ing author. Tel.: +372 5035914; fax: +3727375825.

esses: rein.ahas@ut.ee (R. Ahas), ular.mark@positium.ee

plethora of modelling exercises. These quantitative datasetsand methodologies are standardised nationally and inter-nationally by organisations such as the World Travel andTourism Council (WTTC), World Trade Organisation(WTO) and Eurostat as visitor monitoring and manage-ment tools, and do not raise many methodologicalquestions. Nevertheless, conventional quantitative meth-ods are too limited and restricted to answer complicatedquestions about international tourism flows in a globalisingworld. For example, the traditional statistics on touristflows, such as border and accommodation statistics, do notprovide researchers information concerning the choice ofdestination or the evaluation of objects of interest and theinfrastructure visited. Also, in many European Union (EU)member states as in Estonia, border statistics are no longerrecorded. Accommodation statistics often have problemswith tax violations in Eastern European and othercountries, and overnight stays do not show the daily

ARTICLE IN PRESSR. Ahas et al. / Tourism Management 29 (2008) 469–486470

geographical movement of persons. Tourism analyses playa more and more important role in forming tourismstrategies, plans and marketing tools, in which money frompublic administration is very often used. Because quanti-tative tourism databases remain too simple, more and moretourism studies tend to follow qualitative approaches orfocus on very narrow applied studies. This is stimulating acontroversial debate on the role and potential of academicresearch and about methodological approaches in tourismresearch and geography (Casino & Hanna, 2000; Hall &Page, 2006).

Despite the comprehensiveness of qualitative ap-proaches, tourism studies still need quantitative datasetsin order to monitor tourism flows and to perform academicanalysis of trends, perspectives and geography. Recentdevelopments in information and communication technol-ogies (ICT) such as geographical information systems(GIS) and digital databases are advancing surveyingmethods in geography and tourism studies. The GIS andrelevant visualisation methods have many applications intourism studies (Buhalis & Licata, 2002; Frihida, Marceau,& Theriault, 2004; Lew & McKercher, 2006; Wang &Cheng, 2001). One of the emerging subjects in geographicalstudies is connected with mobile (cellular) phone position-ing datasets and location-based services (LBS) (Ahas,Aasa, Silm, & Tiru, 2007; Ahas & Mark, 2005; Ratti,Frenchman, & Pulselli, 2006; Spinney, 2003). Mobilepositioning data has great potential for applications inspace–time behaviour studies addressed in studying tour-ism geography, though there are various restriction andpre-conditions in ICT applications. Nevertheless, mobilepositioning data tends increasingly to complement con-ventional data sources and destination management, as it isbecoming crucial and more important how a destinationutilises its resources.

The objective of this paper is to introduce and toevaluate the applicability of the passive mobile positioningdata in studying tourism. Passive mobile positioning data isstored automatically in the memory files of mobileoperators as billing information, technical references ofhand-over or other relevant logs with the precision ofnetwork cells (Ahas, Aasa, Silm, et al., 2007). There is ahuge amount of geographical information available on thehard drives of mobile operators, which can be used ingeographical and other relevant studies. The paper assessesthe strengths and weaknesses of the method, such as howto access those databases, issues of privacy, sampling andspatial resolution. We also illustrate our discussion withsome examples from Estonian study projects with suchdata.

2. Data and methods

2.1. Specific terminology

Mobile positioning—tracing location coordinates ofhandsets via the cellular network. There are different

methods for location, such as cell ID, triangulationdirection or/and distance from antenna, determined usingradio waves, A-GPS. Different positioning methods areused because of different network standards (GSM,CDMA, 3G) and for different purposes. The rapiddevelopment of mobile positioning began with regulationsto locate emergency calls (US 911 bill).

Active mobile positioning data—mobile tracing data inwhich the location of the mobile phone is determined(asked) with a special query using a radio wave. In order toask the location of certain phones, a special environmentand permit from the phone holder is required. Activemobile positioning is used in emergency calls, ‘‘friendfinder’’ and many other LBS applications.

Passive mobile positioning data—automatically stored inthe log files (billing memory; hand-over between networkcells, Home Location Register, etc.) of mobile operators(Ahas, Aasa, Silm, et al., 2007). The easiest–billing memoryis recorded when a person uses a mobile phone (callactivity). Passive mobile positioning data is normallycollected with the precision of network cells. For thecollection of passive mobile positioning data, mobileoperators can aggregate anonymous geographical datafrom log files, ultimately not violating personal identityand privacy, and researchers can use it in surveys forscientific purposes. Issues of privacy and surveillance arevery important aspects of this data, and are discussed laterin this paper. Passive mobile positioning data has beenused in many transport and urban studies. The informationon the crowdedness of network cells is used for research,planning or traffic management. Several applicationsdisplay visualised network information in the handsetscreen.

Mobile tracing—the location of a mobile phone isdetermined with special queries with determined timesequence using a radio signal.

Social positioning method (SPM)—analyses of the loca-tion coordinates of mobile phones (tracing or passive data)and the personal (social) features of phone holders forspace–time movement analyses (Ahas & Mark, 2005).

Base station—the cellular network is based on basestations. A base station usually has one tower and severaldirected antennae in the tower. The radio coverage of manyantennae forms a cellular network.

Cell ID—every network cell in a mobile phone networkhas a unique ID, and the location of a phone in the cell caneasily be determined for every call activity. The location ofthe cell is normally determined with the base station whichhas one antenna (omni-round radio coverage) or severaldirected antennae (radio coverage sector, determined withbearing degrees).

The size of a network cell is not fixed; the phone normallyswitches to the closest antenna or that with the strongestradio coverage. If the network is crowded, the phones canbe switched not to the nearest station but to any other inthe neighbourhood. The optimal distance from handset toantenna in the GSM network is less than 60 km. The

ARTICLE IN PRESS

Table 1

Key features of the mobile positioning database used in the study from

April 1, 2004 to August 31, 2005

Study period

April 1, 2004–August 31, 2005

Tourists (random IDs) 1,207,677

Call activities 12,788,049

Average call activities per person (ID) 10.3

Number of nationalities in database 96

Fig. 1. Distribution of mobile positioning database entries (call activities)

in the survey period, illustrating seasonal and weekly rhythms; (A) 2-day

technical gap in operators’ databases and (B) tourists from Russia and

Finland celebrating the New Year in the Tallinn area.

R. Ahas et al. / Tourism Management 29 (2008) 469–486 471

antennae facing the Estonian sea coast are more powerfuland cover longer distances.

Call activity—any active use of a mobile phone innetworks: outgoing and incoming calls; outgoing andincoming SMSs; Internet or GPRS services; LBS; etc. Thismeans that if a person has a mobile phone switched onduring a visit to Estonia but never uses it, his/her presencewas not recorded in this database. In many networks,operators send a ‘‘welcome’’ message if you access theirnetwork.

Roaming—contract between operators that allow the useof mobile telephones in other countries than those in whichit is registered; the operator can recognise the country oforigin of the phone and other details for billing purposes.

Depersonalised dataset means that all respondents(foreign phone identities) remained anonymous, and theoperator provided generalised data: country identificationand the coordinates of the cell in which the phone wasused. The nationality of the tourist is determined by thecountry in which the mobile phone is registered.

Random phone ID—randomly given and anonymous IDnumber for every mobile phone used in the network.Allows researchers to recognise the spatial movement ofpersons.

2.2. Data

The database used for the current case study consists ofthe roaming data of the foreign mobile phone call activitiesof Estonian Mobile Telephone, Estonia’s biggest GSMnetwork (EMT, 2007), for 17 months, from April 1, 2004 toAugust 31, 2005, consisting of 12.8 million call activities of1.2 million visitors (phone IDs) from 96 countries. Despitethe more specific tourist typologies and definitions, in thispaper we define tourists as all non-resident, foreign visitorswho have mobile phones and spend some time (notdependent on the number of days or the type or meansof travel) in Estonia. The nationality of the tourist isdetermined by the origin of the country in which the mobilephone is registered. We have used this tourist positioningdatabase before, in studying the space–time behaviour andseasonality of tourist flows in Estonia (Ahas, Aasa, Mark,Pae, & Kull, 2007). In this article, the resident tourist flowsare not accounted and analysed, as it is another broad areaof passive mobile positioning.

EMT’s nation-wide GSM network is the biggestoperator in Estonia, with approximately 60% of marketshare, using the 900MHz frequency band, as well as1800MHz in city centres and 3G in Tallinn since 2005(EMT, 2007). EMT covers 45,227 km2 of Estonian landterritory and a total of 90,000 km2 including the Baltic Seaarea. The network includes around 690 base station sites(network cells) and serves approximately 600,000 mobilesubscribers in Estonia (EMT, 2007). The main features ofthe database of call activities of the foreign mobile phonedatabase used in the current case study and illustrations arepresented in Table 1, and the temporal distribution of the

database in Fig. 1. EMT had technical problems savingroaming data for 2 days in July 2004, causing a visible gapin the dataset (Fig. 1(A)). Fig. 2 presents the spatialdistribution of phone calls during the study period. Itshould be emphasised that the data of two smalleroperators with roaming services in Estonia are not includedin this study.In order to describe tourist flows in Estonia and evaluate

mobile positioning data, data from the Estonian officialaccommodation statistics has been employed on the countylevel (ESA, 2007). For every month of the study period wereceived data about 15 counties regarding the number ofnights in accommodation establishments, organised bycountry of origin. According to the accommodationstatistics, 2.3 million non-resident tourists stayed inEstonia during this 17-month period. Unfortunately thereis no tourism data regarding border crossings since Estoniajoined the EU, and therefore this time-series ended in May2004.

2.3. Methods

Mobile positioning data has some peculiarities, such ascheaper data collection and lower spatial precision than atravel diary or GPS-based records. Several researchers haveexamined tourist flows in specific environments, for

ARTICLE IN PRESS

Fig. 2. Spatial distribution of roaming calls in Estonia: average monthly number of call activities in EMT network or study period 1.04.2004–31.08.2005.

R. Ahas et al. / Tourism Management 29 (2008) 469–486472

example urban locations (Haywood & Muller, 1988),heritage attractions (Beeho & Prentice, 1995), islandenvironments (Fennell, 1996) and crowded environments(Graefe & Vaske, 1987). Nevertheless, the mapping andmodelling of tourist movements is an underexplored fieldof tourism research; there has been no systematic develop-ment of a universal model for the analysis of visitors’space–time flows. Some simplistic models could be visuallyappealing but may lack usefulness and broad applicability.Others are too comprehensive and are stuck in datamanagement and modelling (Lew & McKercher, 2006).Intelligent positioning as a survey methodology in tourismresearch is starting to mature, but it is still quite marginal,with a lot of hindrances and sophisticated ICT solutions.As the mobile positioning method is competing with manyother conventional methods for quantitative research,researchers and tourism industry must begin analysing itsperformance and applicability to destination marketing, inaddition to its technological break-through if it gains widerimportance in tourism research and become a crediblemethodology for serving tourism planning.

Mobile positioning data can be analysed using differentmethods of space–time behaviour and activity spaceanalysis (Hagerstrand, 1968; Kwan, 2000; Timmermans,Arentze, & Joh, 2002). The survey method that uses mobilepositioning data is also called the SPM in Estonia, and thisuses the location coordinates of mobile phones andpersonal features of phone holders for space–time move-ment analyses (Ahas & Laineste, 2006; Ahas & Mark,2005). Unfortunately, the passive tourist positioningdatabase only has a limited number of features for eachtourist: country of origin and a random ID number whichallows recognition of the same tourist during a visit toEstonia. The tourism roaming database is stored in thepositioning server of LBS company Positium ICT in the

PostgreSQL 8.1 environment, which is extended withspatial extension PostGIS. In the current study, the spatialanalyses were performed using ArcGIS 9, the InverseDistance Weight interpolation mode and Kernel Density.

2.4. Tourism trends in Estonia

Estonia is an emerging destination in north-easternEurope, and is well known for the historical heritage of OldTallinn, nature tourism experiences, spa cluster and cheapservices. There are summer and winter holiday destina-tions, a metropolitan hub and destinations for budgettravellers, as well as favourite destinations for ecotourismand sporting (Ahas, Aasa, Mark, et al., 2007; ESA, 2004;Unwin, 1998; Worthington, 2003). The rapid growth since1991 was powered by a demand for the new attractions of apost-Soviet society and a demand for cheap goods andservices by neighbouring Scandinavian countries. Todaythe changes in Estonia (EU membership, rising prices) andin the international tourist market require the use ofcompletely new strategies to overcome difficulties intourism services and the unbalanced demand and supplycaused by growing global competition.In 2002, 1 million tourists visited Estonia, and in 2005

1.45 million tourists were accommodated (ESA, 2007). Theestimated total number of foreign visitors was 3–4 millionpersons. In 2004, 66% of non-resident tourists in accom-modation establishments were holiday tourists, 5% wereconference tourists, 18% were other business tourists and11% were tourists travelling for other purposes such asvisiting spas (ESA, 2007). The origin of tourists haschanged somewhat from 2004 to 2006, i.e. after joiningthe EU. The number of Finns, who are the predominantgroup (50–60% of total tourists), has decreased by 5%, and

ARTICLE IN PRESSR. Ahas et al. / Tourism Management 29 (2008) 469–486 473

the proportion from other European countries has risenaccordingly.

The share of tourism in Estonia’s gross domestic product(GDP) at current prices was approximately 8.2% from1997 to 2000 (ESA, 2004). Based on growth predictions,tourism is destined to continue playing a vital role inEstonia’s economic and social development. The mainchallenge of Estonia’s tourism policy will be to create theprerequisites for an adaptation to the new situationprevailing in international tourist markets. The mainemphasis was not on efficient marketing, but fulfilling theneeds of a booming market. These features of an emergingtourism market with changing tourism flows open a goodexperimental environment for the testing of mobilepositioning methods in the Estonian tourism market.

3. Accessing passive mobile positioning data

3.1. Restricted access to data

The biggest problem involved with the use of mobilepositioning is how to get access to data. This problem isquite similar for active and passive mobile positioning.Limitations are related to many important aspects. Mobilecommunication is a field of active competition betweenoperators, hardware and software providers and otherrelated businesses. Because of business secrets, the opera-tors conceal all data and numbers that may provideinformation to other operators. This is why in mostnetworks the location of antennae is secret, even if peoplecan see them every day. Subscribers’ privacy and surveil-lance concerns are also an important aspect limiting the useof data by third bodies. Operators do not want to loseusers’ trust, and even a rumour or unsuccessful pressrelease about research in an operator’s database may causedamage to the operator.

Another important aspect limiting the use of this type ofdata is that the chain of the value of LBS is too long. Tosucceed, LBS or studies with positioning data need to becompetent in several important fields: (a) to access theoperator’s database; (b) to cope with the peculiarities ofmobile operators’ hardware and software; (c) to work withhuge databases; (d) to handle data security; (e) to befamiliar with GIS, geographical data and statistics; (f) tohandle social sciences methods; and (g) to address theneeds of end-users (academic or applied). It is a necessityfor very profitable business plans to cover all of the diversechain of values described here. Most LBS applications,however, have a normal price, and it is difficult to cover thecosts of the whole chain of value. Mobile operators veryoften use the ‘‘revenue share’’ business model for LBS, andtheir ‘‘appetite’’ is actually one major limiting factor in thesuccess of LBS. The LBS and relevant research projectstoday need corporate institution or close cooperationbetween organisations to cover all of the chain of values.The problems of not developing LBS instead of the boomin early 2000 are related to the same issue—the length of

the chain of value and related costs. Very often LBSapplications are developed by mobile operators, but theydo not know the needs of end-users and specific GIS tools.The result is that we can see the number of LBS offered bymobile operators but they are not practical or comfortable.Today, Estonian mobile operators offer about 10 LBS-related services, the most successful of these being thetracking of expensive cars for case of theft. Another seriesof positioning applications is developed by end-users (alsoacademic), but they lack knowledge of operators’ hardwareand software and have difficulties with access to data.Therefore the successful development of positioning-basedapplications and research has been successful in a fewbigger institutions, for example MIT (Ratti et al., 2006) orsmall countries with good cooperation between institu-tions, for example Estonia (Ahas, Aasa, Mark, et al., 2007).Good relations with the mobile operator are not the only

factor determining the availability of passive positioningdata. It is also necessary to guarantee secure communica-tion channels and IT solutions, servers and knowledge ofhandling spatial data and a good GIS analysis, as simplelocation coordinates are not a product for end users. Forexample, for successful data handling in the Estonian case,the Institute of Geography of the University of Tartu isdeveloping a cluster of positioning servers together withLBS Company Positium ICT (Positium, 2007).

3.2. Concerns of privacy and surveillance

Privacy, surveillance and data security are the keyproblems in using passive and active mobile positioningdata. The development of fears concerning privacy and theaccompanying debate in society is uncertain. The growingICT generation may be open to all that is interesting andnew. At the same time, the fear of terrorists or anoveractive ‘‘hunt for terrorists’’ may seriously jeopardisethe development of ICT as an electronic advertisementbusiness. Privacy is not only discussed by the subjects ofthe possible surveillance, it is a much broader publicconcern. There are fundamental questions of personalfreedom and scientific ethics that cause public debate inthis matter. Instead of legislation and public debate, thekey question today is that mobile positioning data is veryprivate, and operators do not want to violate subscribers’trust.The legal aspects of data collection today in Estonia

regulate that all data collection must be performed inaccordance with European regulations (DIRECTIVE,2002/58/EC) and Estonian national regulations (EAI,2007) about electronic communications and personal datahandling. This is the case for both active and passivepositioning. Mobile operators follow this aspect verycarefully. Data intermediate Positium ICT must secureboth sides, as operators and end-users want to be sure thatdata is secure and privacy issues do not jeopardise services.The passive mobile positioning dataset we use in this

study is very carefully compiled by EMT, the largest

ARTICLE IN PRESSR. Ahas et al. / Tourism Management 29 (2008) 469–486474

operator in Estonia. The data protection regulation isfollowed in all aspects: persons or their locations cannot beidentified. There are special aspects in the database optionsarising from the Data Protection Act. For example, allrecords with less than 5 entries of one nationality in onenetwork cell were presented as ‘‘other’’ countries, and theirproportion was very small (less than 0.1%). The persons(phones) are marked by operators with random IDs such asFIN No. 0000001, which cannot be linked to any personalidentification. EMT has a long-term contract with posi-tioning company Positium ICT, which stores the datasetand prepares queries for end-users such as scientists.

In a technical and legal sense, the data is managed well,but there are more important questions to discuss at thispoint. Nevertheless, people do not feel good aboutpositioning and surveillance issues. The data is almostsimilar to other statistics gathered from hotel records suchas accommodation statistics. But psychologically, thepositioning data is different. Mobile phones have becomevery personal and there is an unconscious fear that one canbe traced and listened to everywhere one goes. Ourexperiences with active mobile positioning in Estoniaduring 2004–2006 show that people are becoming accus-tomed to positioning and are not as concerned as theyinitially were. Later they became interested, and every thirdperson even volunteered to participate in more of ourstudies (Ahas & Laineste, 2006). It is, however, necessaryto explain all aspects of the study and to demonstrate theresults. Explanations and results help reduce fears ofviolation of privacy.

Another question is whether positioning and fear to bepositioned can influence a tourist’s choice of destination. Ifpeople are very concerned about privacy or to bepositioned they do not go to Estonia where tourists are‘‘monitored’’ via phones. The media can magnify this fearand this is one side effect of using such innovations. Thisimpact on travellers can be similar to locations with toomany questionnaire surveys or overly ‘‘friendly’’ voluntarytourist guides in some exotic country.

All privacy issues are special, however, and must beaddressed very carefully. The use of passive positioningdatasets will probably need special discussion and regula-tion in the future, in order to make all stakeholders happy

Table 2

Original log statement of the mobile positioning database

Time Rand. ID Cou

April 1, 2004 00:00:03 510193 Lat

April 1, 2004 00:00:04 43204 Fin

April 1, 2004 00:00:23 55861 Fin

April 1, 2004 00:00:24 770841 Rus

April 1, 2004 00:00:31 505262 Lat

April 1, 2004 00:00:35 49104 Fin

April 1, 2004 00:00:36 586313 UK

April 1, 2004 00:00:39 585543 UK

and leave the option of being ‘‘excluded’’ from surveys orgeneralised to a safer level.

3.3. Gathering passive positioning data

As neither researchers nor any other persons can accessmobile operators’ strictly protected databases, the databasewith passive positioning data is typically delivered byoperators. In some cases the cooperation between opera-tors and LBS applicants is organised as limited access to apositioning server. Estonian data is gathered by thecompany Positium, which has a special contract with twomajor mobile operators for intermediate data and toguarantee data security (Positium, 2007). Positium haslimited access to request data and to download it. Foractive positioning (tracing individual movement) is devel-oped authorisation tool in combination with SMS message.Some of Positium’s services are visible online from thepositioning server www.positium.com, for visualisationactive positioning data (tracing experiments and games)in real time and for passive positioning (tourist and localcall activities) in semi-real time.The tourists’ passive positioning database for the current

case study included the following features for every callactivity: (a) the time of the call activity; (b) the randomlygenerated ID number of every phone used as the ID of onephone (person); (c) the country code of the phone (person);and (d) the cell ID with the x and y coordinatesof the antenna. An example of the database is presentedin Table 2.

3.4. Spatial resolution of the method

The spatial accuracy (precision) of the location informa-tion in the passive positioning database depends on thedensity (quality) of the mobile network and positioningmethodology. Different producers of network infrastruc-ture, such as Nokia or Ericsson for example, have differentstandards for network architecture (size of cells, coverage,and type of antennae) and different positioning methods(Ahas, Laineste, Aasa, & Mark, 2007).

ntry x y

via E24-52-24.00 N59-26-12.00

land E24-45-34.00 N59-26-36.00

land E24-44-37.00 N59-25-49.00

sia E27-44-38.00 N57-52-12.00

via E24-45-34.00 N59-26-36.00

land E24-44-41.00 N59-25-27.00

E24-43-53.00 N59-26-06.00

E24-44-37.00 N59-25-49.00

ARTICLE IN PRESS

Fig. 3. Distribution of EMT base stations and relevant network cells in Estonia, with Voronoi polygons (polygons whose boundaries define the area that

is closest to each point relative to all other points; they are mathematically defined by the perpendicular bisectors of the lines between all points).

R. Ahas et al. / Tourism Management 29 (2008) 469–486 475

3.4.1. Network density

Generally, the network is denser and the locationinformation is more accurate in urban areas with densenetwork coverage. Estonian network of EMT is presentedin Fig. 3. As population density and transportationfrequency (the number of possible calls in an area) is amajor determinant in network planning, the structure ofthe mobile network actually follows the structure of localcommunities and highways. In rural areas, the network andantennae are directed along busy highways and towardscrowded city centres. In Estonia’s biggest cities such asTallinn, Tartu, and Parnu, one can expect locationaccuracy with active mobile positioning of 100 to 1000m.Suburban areas vary between 450m and 2 km (Ahas &Laineste, 2006). Due to the sparse nature of the operatorbase stations in rural areas, accuracy can vary from 1.5 to20 km (in Estonia’s remote wetland areas). The larger errorin rural areas is a problem in studying visits to objects witha great level of detail. In most cases in Estonia andprobably in other countries as well, the less-populated ruralareas have less tourists, and therefore the impact of spatialerror on overall data quality is smaller.

3.4.2. Interpolation of network data to administrative units

One standard problem in the application of positioningdata is that the network cells and actual distribution of callactivities recorded do not geographically match theadministrative units. The main users of tourism surveysare local communities, parishes or counties, and theynormally want to have data generalised into administrativeborders in detail. This is also a question in the parallel useof other statistical databases that are normally generalisedto administrative units. In order to fit the detailed cell(antenna) information with administrative units, the 690

antennae (network cells) were summarised for 224 munici-palities and 15 counties. The generalisation model byadministrative units might not be the most suitable,because the antennae and relevant geometrical cells andactual distribution of call activities do not precisely matchadministrative borders. We used the technique where thecell (antenna)-based information was simply interpolatedfor the territory of municipalities. This solution was notideal, however, as the number of total call activities in anarea should in total be constant. Therefore we more oftenuse the level of network cells for detail studies, or countiesas larger territorial units and therefore correspond moreclosely to the network cells. At the present time ourworking group uses three scale levels for analysis: (a)counties; (b) municipalities; and (c) network cells repre-sented using Voronoi polygons (Fig. 3). Every level has itsadvantages and disadvantages depending on the objectivesand scale of the study.

3.4.3. Spatial noise

The passive mobile positioning data has several bias andspatial noise. One aspect of spatial noise appears in borderand marine areas. Ships and passengers on the sea may usenetworks that lie relatively far from the coast (amplifiedantennae in coastal areas), and they might never visitEstonia. Some base stations on the seaside (in north-western Estonia) are much overburdened in our databases.The most likely reason for this is calls from passing ships.A similar cross-border impact is also possible in con-tinental areas as people from cross-border areas may useEstonian mobile network cross border. The cross-borderuse can be accidental or planned because of pricedifferences etc. Both cases of cross-border noise needattention from end-users of data. In some cases we

ARTICLE IN PRESSR. Ahas et al. / Tourism Management 29 (2008) 469–486476

eliminated questionable network cells from analyses. Inother end the tourism industries in border areas wereinterested about potential demand nearby. Even if passingships never reach land, this information is valuable formany tourism studies, because we can learn about nearbytransit and plan strategies with which to benefit from thispotential. The impact of the hand-over of calls between theclosest network cells is also a problem in inland areas.Moving and standing people/phones are switched (handedover) to different antennae, and the call is not fixed to theclosest or strongest antenna. If one antenna is crowded orvisual contact is disturbed, the phone can be switched toanother nearby or distant antenna. GSM coverage isnormally not more than 60 km horizontally. This fact hasto be taken account if we use positioning data orinterpolation models.

There are other important aspects of the geographicalinterpretation of data connected with directed antennae,elevation models or base maps for data analysis. Forexample, in studying subjects who are driving, it isimportant to know that even if an ongoing call is handedover to other antenna, the location of the call activity is stillrecorded as the place where the call started.

3.5. Sampling and representativeness of the data

Our experiences in presenting passive mobile positioningdata and studies to other scientists and end users show thatcritical attention is paid to the sampling and representa-tiveness of the data. This data source is new, and there is aneed for future studies concerning samples and sampling.For example, the passive mobile positioning database usedin Estonia consists of a full sample of foreign phones usedin the EMT network. The digital database has a hugenumber of entries: 12.78 million locations and 1.2 milliontourists (phone IDs). Using powerful database manage-ment systems and analytical tools, it is possible to studyinteresting aspects of tourism geography. But there areseveral questions to address in this matter concerningrepresentatives of this dataset.

Fig. 4. The share of tourists (IDs) and the number of call activities for the

12 most active visitor countries in the Estonian mobile positioning dataset.

Average number of call activities per ID.

3.5.1. Penetration of mobile phones and roaming

We do not know the exact proportion of phones andphone use between nationalities and social groups in theirhome country or travelling. Therefore we can say that weare studying the geography of foreigners in Estonia whouse mobile phones registered abroad (roaming). With thelatest Eurobarometer survey (2006), we know that theaverage number of mobile phone users has reached 79% asan average for the EU; 90–95% of the citizens ofLuxembourg, Sweden, Finland and the Netherlands usemobile phone. The number is smaller in Poland, Hungary,Greece, and Spain, with an average between 64% and75%. In Estonia 86% of citizens use mobile phones(Eurobarometer, 2006). Global differences in phonepenetration are small because developing and transitional

countries have very progressive numbers of mobile phoneuse.It is also true that different social groups in one nation

use phones differently. For example, it can be surmisedthat businessmen and officials use mobile phones morethan pensioners, because of their finances and activebusiness life. The younger the respondents, the more likelythey are to have a mobile phone. A large number of seniorsalso own a mobile phone: more than one in two Europeansaged 55 and over have a mobile phone (Eurobarometer,2006). The habits and ‘‘mobile culture’’ of certain groups oftourists may also have an influence. Many groups ofGerman pensioners visit Estonia whom tour guidesconsider not being active mobile phone users. At the sametime, many persons who do not use mobile phones at homein everyday life take phones along when travelling, so thatthey can keep in contact with home. In addition, manyvacation tourists might consciously not want to ruin theirvacation with phone calls, so they switch off their phones inorder to forget the worries of work and home when theytravel.It is clear that different user groups and nationalities

make different number of call activities during a day. Animportant indicator that characterises the mobile position-ing data used is the number of call activities per visitor(ID). The average number of call activities per ID (person)in Estonia was 10.3 during the entire survey period. Thehighest numbers were for Latvians, Danes and Norwe-gians—up to 20 call activities by ID; the smallest numberswere for Spanish, Italian and Austrian visitors, 5–7 callsper ID. This figure identifies the mobile behaviour andculture of visitors. The comparison of the share (%) of callactivities of the top 12 countries and the share ofcorresponding IDs shows that both call activities and IDsdescribe the number of tourists relatively well, andvariation remains within the range of 0–3% (Fig. 4). Onthe basis of that, we selected for better indicator to studysingle-call activities, which if necessary can be generalisedto the pool of IDs and relevant activity spaces.

ARTICLE IN PRESSR. Ahas et al. / Tourism Management 29 (2008) 469–486 477

3.5.2. Different network standards

There are different standards for cellular phones indifferent parts of world. Tourists from American or Asiancountries use phones with other standards (CDMA andothers), which are not compatible with the GSM networksused in majority of Europe and in Estonia. Thereforevisitors from some countries can be underrepresented in thepositioning database. For example, tourist from the USA(CDMA network) hold 10th position in Estonia in terms ofaccommodation statistics, with 1.5% of the total, and withpositioning data their position is 23rd, with 0.3%. For anew and expensive telephone handset or 3G networks,compatibility with other networks is a smaller problem.For people using cheaper phones this is a dual problem:foreigners in Estonia and Estonians in those countries.

3.5.3. When do tourists use phones? Can this change the

distribution of location points?

This aspect is very important, because some touristtypes, for example business travellers, leave more trackingpoints (call activities) during the work day, and sometraditional tourists, for example those on bus excursions,make calls only during the evening hours. We studied thediurnal distribution of tourists’ call activities as an hourlyaverage of the entire study period (Fig. 5). The averagenumbers and the curve of tourists’ calls is similar onworkdays and on the weekend: calls start after 7AM andreach peak frequency after 10AM, whereas after 3 PM thenumber of calls slowly starts to diminish until midnight.On weekends there is still active calling from midnight to4AM, while on workdays there are fewer calls aftermidnight. Actually, the distribution of calls by Estonians inEstonia is quite similar, only the daytime weekend periodhas 30–40% less calls than on workdays.

We studied the geographical differences between callsmade during the day and at night. The distribution waspredictable, as daily calls are spread all over the country,although of course the majority are still in major cities andtransit highways. The evening calls (after 9 PM) are

Fig. 5. Distribution of hourly average sums of tourists calls for workdays

and weekends during the study period, 1.04.2004–31.08.2005.

concentrated more and more in major cities, and the restof Estonia is less thoroughly covered.One aspect that may influence the distribution of calls is

that many cellular networks send ‘‘welcome’’ messageswhen new phones enter the roaming network. This is a veryvaluable service that enables us to obtain first call activityfrom the place of entry. The EMT network does not sendsuch a welcome message, but our analysis of the distribu-tion of first call showed that many phones get first callactivity at the airport (Fig. 2). Other border crossing areasalso have a high number of first calls, but distribution ismuch more dispersed, as radio coverage starts before theborder, and hand-over between operators can be delayed intime or distance.

3.5.4. Does the pricing policy in mobile services influence the

use of phones abroad?

Different nations also use phones differently because ofthe different costs of roaming calls and the different incomelevels in their home country. This is an important aspect wemust remember when using this kind of data. For theaverage Ukrainian, roaming in Estonia is much moreexpensive than for Germans because of the standard ofliving and because of different prices. An average of 80%of Europeans have a mobile phone, and nine out of tenusers use roaming services when travelling abroad (Euro-barometer, 2006). A lower proportion of Europeans do notuse their mobile phone abroad: 8% switch it off and 7% donot take it with them. The Eurobarometer survey alsodemonstrated that excessive communication costs are themain (81%) reason why Europeans use their phone lessoften when travelling abroad. On average, roaming pricesare still four times higher than domestic mobile calls.Businessmen use roaming services much more than regulartravellers.

3.5.5. Other aspects concerning sampling

Cross-border noise or ‘‘hand-over’’ between mobilenetworks described earlier here is also a sampling problem.People living in border regions often use manual networkselection, because it is expensive to be accidentally switchedto the network of another country and be charged‘‘roaming prices’’.Another error in the data of call activities could be

related to the length of the visit. It is not easy to distinguishwhether the person left the country and returned or stayedthere all the time. Use of phones can be also influenced bypossibilities for charging phone batteries, for examplepeople hiking in less habited areas may have such aproblem. To compare mobile tracing experiments withGPS, the mobile positioning has great advantages, aspeople really care about battery of mobile phones.The quality of radio coverage can also have an influence

on sampling. Estonia is a small country, and all threeoperators have very good radio coverage over 90% of itsterritory. In many countries radio coverage is not total, andpositioning may have gaps which influence products or

ARTICLE IN PRESSR. Ahas et al. / Tourism Management 29 (2008) 469–486478

research outcomes. Sample is influenced also by the factthat operators belong to international cooperation net-works which direct their clients to a certain network in aforeign country also has influence. This impact was notdetected in the current database (comparison with accom-modation data) of EMT whose partners are fromScandinavian countries. This may be because of the largeamount of data in currents database which minimiseseffects caused by the impact of preferring certain operators.

We must consider all aspects that may influence thequality of data. After using passive positioning data inseveral studies and comparison with other data sources, weare optimistic. We believe that our database describes

Fig. 6. Distribution of call activities by Estonian counties during the 17 mo

nationalities during the study period.

Table 3

The number of non-resident tourists in the mobile positioning and

accommodation data of the top 12 nations for the period

01.04.2004–31.08.2005

Country Call activitiesa % Accommodation nightsb %

Finland 6,803,493 51.5% 1,305,433 56.6%

Latvia 1,064,015 8.0% 70,053 3.0%

Russia 843,156 6.4% 67,193 2.9%

Sweden 811,384 6.1% 156,126 6.8%

Lithuania 641,665 4.9% 45,232 2.0%

Norway 558,980 4.2% 58,632 2.5%

Germany 509,884 3.9% 169,304 7.3%

UK 365,196 2.8% 78,689 3.4%

Poland 261,548 2.0% 21,625 0.9%

Denmark 204,225 1.5% 21,982 1.0%

Italy 202,430 1.5% 46,111 2.0%

Netherlands 121,036 0.9% 19,218 0.8%

aEstonian Mobile Telephone data.bEstonian Statistical Board data.

‘‘foreign visitors who use mobile phones in Estonia’’, andthe quantity of the data allows us to say that it is arelatively good sample for describing the majority oftourists. We need to study phone use routines and statisticsin greater depth.

4. Space–time flows of foreign tourists in Estonia

In order to demonstrate the applicability of mobilepositioning data in tourism research, we present someresults of the geographical case study in Estonia. The originof non-resident tourists by country during our surveyperiod in two alternative databases, positioning andaccommodation statistics used in this study, is presentedin Table 3. It is quite natural that the greatest numbers ofvisitors to Estonia were Finns and people from otherneighbouring countries in the Baltic Sea region, as well aspeople from larger countries.

4.1. Geographical distribution and flow pattern of tourists

4.1.1. Spatial statistics

Spatial distribution of tourist flows and regionaldifferences in Estonian tourists’ share at the county levelare presented in Fig. 6. On the basis of geographicalstatistics, it is possible to analyse densities, trends, spatialcorrelations and clustering, and to model space–timebehaviour. The most interesting results from county leveldescriptive studies were connected with regional changes intourist flows and the share of nationalities after joining theEU in May 2004. The number of Finns began to decrease,and that of other EU nations to grow. The presence ofFinns remained high in Tallinn and Western Estonian cities

nths studied: total number in thousands and share of 5 predominating

ARTICLE IN PRESSR. Ahas et al. / Tourism Management 29 (2008) 469–486 479

with beach and spa attractions. In inland areas and smallercities, the number of Western Europeans rose, and alsonumber of Russians increased in the eastern part ofEstonia.

We analysed seasonal and weather-caused variabilityin tourists’ movements in Estonia with this database(Ahas, Aasa, Mark, et al., 2007). Seasonality createsdifferent tourism patterns in Estonia: summer is orientedto western Estonia and Islands, winter to cities and inlandareas. Mean diurnal air temperature causes significantchanges in tourist flows in coastal areas and in some inlandsites during the summer months. Tourist flows in someareas such as bigger cities (Tallinn, Tartu) do not dependon the weather or seasons, they have clearly distinguishableweekly rhythm empowered by business traveller’s max-imum on Wednesday and weekend visitor’s maximum onSaturday. Also the highway corridors have clear weeklyfluctuation of tourist flows (Ahas, Aasa, Mark, et al.,2007). It is also possible to model space–time behaviourusing travellers’ IDs, for example to detect where theymade decisions because of changing weather, and topredict future movements. Such a model with air tempera-ture data was generated for the behaviour of tourists on theParnu and Kauksi beaches, and the model can predict upto 80% of summer movements in beach areas.

Estonian local municipalities use general tourism statis-tics for the compilation of development plans and to maketourist development funding applications for the adminis-tration of EU programmes more trustworthy. One inter-esting use of municipality statistics is connected withcompetition: despite having their own statistics, munici-palities want to know and compare how tourism isdeveloping in close or competing areas. On the networkcell level, the mobile positioning data is used to studytourism in urban space for master planning projects.Network cell data was also useful for the study ofgeographical distribution and the impacts of single tourismevents such as local festivals. Network cell statistics alsohelped to discover some popular visitor areas inEstonia that were not recognised properly, for instanceLatvian fishermen’s spots on the coasts of Lake Peipsi in

Fig. 7. Location of the first call in Esto

winter weekends; or the space–time behaviour of richRussian New Year’s eve tourists in the Estonian capital,Tallinn.

4.1.2. Activity spaces

The random/anonymous ID of phones allows us torecognise the movement of one tourist during his/her stayin Estonia or to recognise in case of return trip to Estonia.This opens many interesting study perspectives andapplications. As we can study the movement of one personover a certain time, we applied methods of analysingactivity spaces as studying routes, density of visited sites orgeometrical distribution of visited places. A digital move-ment track of a person could be compared to a DVDmovie—we can study visits to a certain moment or place,and we can rewind (track) it or fast forward it to study allthe places visited. We have studied one very importantgeographical feature of tourism—the time and place ofentry into the country for different visitor groups—bymapping the ‘‘first call’’ in Estonia. Fig. 7 shows thelocation of first call activity in Estonia for Latvians andRussians during the study period. The maps present a cleardifference in the geography and in the direction ofmovement of Latvians and Russians. Disaggregating layerby layer, it is possible to study points of entry (ordeparture) in connection with certain seasons, events ordestinations. This is valuable information not only forgeographers but also for marketing or informationservices.The digital movement track also offers possibilities to

study the routes of visitors detected in a specific area and/or time. For example, we analysed the behaviour of visitorsto the International Hanseatic Festival in Tartu: where andwhen visitors entered the country; how long they stayed inEstonia, and which other places they visited here and(overnight) accommodation locations in the Tartu region.The festival’s organisers in Tartu will use this data to planthe next festival and determine how to share costs or betterplan services with other municipalities that also benefitfrom visits by festival members.

nia (A) Latvians and (B) Russians.

ARTICLE IN PRESSR. Ahas et al. / Tourism Management 29 (2008) 469–486480

4.1.3. Linking destinations

The tracks of visitors left in roaming databases can belinked and modelled along street/highway canyons formapping tourism routes. The generation of route maps ofpopular destinations for different nations or locations can beused as valuable background information for tourismstrategies. For example, the linear presentation of touristmovements gives an idea of flows of Latvian and Russiansduring the midsummer holidays, from 22 to 25 June 2005(Fig. 8). The midsummer holiday is different from regulardistribution of people, because events take place in ruralareas. The majority of movements departing from or arrivingin Tallinn, the metropolitan hub, has the harbour andinternational airport, and as a result is the focus of majorinternational transit flows. Latvians dominate in theTallinn–Riga axis and Russians in Tallinn–Narva axis. Theother important tourism locations such as Parnu and Tartualso have links with the major Estonian districts. A few daysafter Midsummer’s Day, most tourists are back in thesummer tourism capitals Tallinn, Parnu and Kuressaare, andthe rest of Estonia is relatively devoid of tourists.

4.2. Monitoring tourism in remote areas and transit corridors

Passive mobile positioning data can also be a veryeffective method for the investigation of tourism in less

Fig. 8. Linear movement corridors of (A) Latvians and (B) Russians in

Fig. 9. Finnish and German tourist records in the remote Soomaa national par

study period.

visited rural and wildlife areas. Those areas often presentspecial problems in terms of tourism development andsettings, owing to the lack of product, market access,shadow economy, infrastructure and problems with carry-ing capacity. The application of traditional questionnairesor counting is a relevant method in places with manytourists, where contacts are easily reached. In less-visitedplaces such as natural parks, individual landscape objectsor small towns, it is very difficult to obtain a bulk ofinterviews, as it is difficult to find those rarevisitors, in particular if there are not designated entrypoints and gateways. In these places, mobile positioninginformation can be the only reasonable source foranalysing the presence of tourists, destinations and theroutes of rare visitors. Fig. 9 gives visitors’ dynamics in aselected ‘‘white spot’’ in the Soomaa wetland protectionarea in western Estonia. As we can see, there are still aregular number of visitors in this remote area. Theplace is popular among ‘‘western’’ visitors, since ourclosest neighbours, for instance Latvians or Russians, donot come to the Estonian wetlands. The visitor profileprovides important information for the planning oftourism development or environmental management insuch specific areas. A map of ‘‘white spots’’ can be animportant data source for tourists who avoid crowdedtourist attractions.

Estonia during the Midsummer’s Day holidays (June 22–25, 2004).

k, wetland reserve. Total of 720 call activities by foreign tourists during the

ARTICLE IN PRESS

Fig. 10. Detected transit tourists in three profile cuts on Tallinn–Tartu highway.

Fig. 11. Average weekdays of visits to Estonia by citizens of Great Britain

(black) and Finns (grey), April 1, 2004–August 31, 2005.

R. Ahas et al. / Tourism Management 29 (2008) 469–486 481

As it is difficult to find the rare visitors for interviews inless-popular areas, there is also a similar problem reachingand describing tourists on big transit channels such ashighways, railways or the sea. High volumes of tourists areregistered along transportation corridors. It becomes clearthat they do not stop very often on the way. Passivepositioning data is a method that allows us to estimate thenumber and profile of tourists passing by in cars or trains.We present here three profile cuts on the Tallinn–Tartuhighway, which is a major transportation zone in Estonia,connecting the two biggest cities (Fig. 10). The number ofcars passing via profile points is biggest in Kose nearTallinn (No 1). There is a recorded average monthlynumber of tourists 33,818 with majority (64%) of Finnsand 6% of Swedes. The second number of transit tourists isdetected in Laeva near Tartu (No 3.) with 10,350 tourists.Number of Finns is smaller (48.4%) and second place haveRussians with 9.4%. The Poltsamaa profile (No 2.) hasvalues between Tallinna and Tartu. This information canhelp estimate the potential demand for certain services andtourist products. In future, transit tourists can be reachedand personally addressed with intelligent signs alonghighways. Intelligent signs will recognise tourists’ nation-alities via positioning data, and automatically switch thelanguage or relevant information on the signs.

4.3. Temporal distribution of tourists

Tourists’ activities in every country have a specificrhythm, i.e. diurnal, weekly or seasonal cycle. There aredifferences between the temporal rhythms of holiday,

weekend and business travellers, as there are between closeneighbours and distant visitors. For example, the Britishmostly visit Estonia on weekdays (peaking on Wednesday),while Finns enjoy Estonia on the weekends, peaking onFriday (Fig. 11). Friday is the busiest day for the majorityof tourists. The distribution of visit days enables one torecognise the purposes of the visit, i.e. business or vacation,and to determine the favourite season and days. On thetemporal axis, the survey was able to filter all four majorsegments—tour groups, special interest visitors (hiking/shopping/spa tourists), independent travellers, and busi-ness travellers.The temporal dimension of tourist behaviour also makes

it possible to estimate the time spent in a study area. We

ARTICLE IN PRESS

Fig. 12. The dynamics of tourists’ (eight dominant nations) call activities during the winter holidays in Tallinn.

Fig. 13. Monthly comparison of accommodation data and call activities.

R. Ahas et al. / Tourism Management 29 (2008) 469–486482

studied the length of the stay of tourists in Tartu. Thelength of stay corresponds to the distance from the homecountry: people from more distant countries stay longer inEstonia than do their close neighbours. For example, themajority of Finns and Latvians stayed 1–2 days; Germanand British visitors 3–4 days; Italians and Americans evenlonger.

The temporal profile of tourists’ movements has greatapplicability. It allows the examination of the impacts oftourism events, to monitor the efficiency of marketingcampaigns and to share costs with benefiting municipa-lities. Fig. 12 shows the distribution of tourist calls inTallinn during the New Year period. Finnish touristscrowd into Tallinn before Christmas (seasonal shopping),whereas wealthy Russians stay in Estonia near/after theNew Year. The overview of the temporal and spatialdistribution of celebrating tourists in Estonia was usefulinformation to show that money spent on the advertisingcampaign in Russian cities was effective. The problem withcelebrations such as New Year’s Eve is that call activitiesare not proportional, due to the prevalence of Happy NewYear congratulations.

4.4. Comparison of mobile positioning data and

accommodation statistics

In order to use mobile positioning data in tourismstudies, it is necessary to validate its quality. The mobilepositioning dataset was compared with traditional accom-modation statistics collected by the Statistical Office ofEstonia. As shown in Fig. 13, the average monthly numberof call activities in EMT network was 780,167, which was 2times higher than the average monthly number of

accommodation nights in Estonia, 386,837. The differencebetween accommodation and mobile data is greater insummer and smaller in winter, and differs between nations.Estonia’s closest neighbours, the Finns, Latvians, Russiansand Lithuanians, have a proportionally greater number ofrecords in mobile positioning data (both call activities andIDs) than the nights spent in Estonia (Table 3). Latvians,for example, have an average share of 8.0% in Estonianpositioning data, and only 3.0% in Estonian accommoda-tion data; Russians accordingly 6.4% and 2.9%; Lithua-nians 4.9% and 2.0%. More distant countries such asGermany have opposite share: 3.9% in positioning dataand 7.3% in accommodation data; Italy 1.5% and 2.0%.Nevertheless, our study shows that the differences are notso great, and several expert opinions say that positioningdata may be more reliable because accommodation

ARTICLE IN PRESSR. Ahas et al. / Tourism Management 29 (2008) 469–486 483

statistics have weaknesses. For example, the smaller andless-expensive establishments do not register all visitors dueto tax avoidance, which is a issue in the rural tourism, andnot only in Eastern Europe.

The correlation between monthly accommodation andpositioning statistics for all 15 counties during the studyperiod is very high: 0.99 (Po0.05). Assessing correlationsby county (Fig. 14), the correlations are high in 8 counties,with coefficients of more than 0.9, reaching the maximumin high tourism counties such as Parnu (0.99), Saare (0.99)and Laane (0.98), as well as in the tourism-oriented Tallinnarea, Harju (0.96). Lower correlations between accommo-dation and mobile positioning statistics are found incounties with less developed tourism industries, such asJogeva (0.65), Voru (0.68) and Valga (0.77).

In order to evaluate the quality of mobile positioningdata, comparisons between accommodation and mobile

Fig. 14. The correlation of monthly sums of acc

Fig. 15. Distribution of accommodation statistics and call activities i

positioning statistics were studied in greater detail in Laaneand Jogeva counties. In the case of Laane County, thecorrelation between the mobile positioning data andaccommodation statistics was very high (0.98), with thetwo datasets almost identical (Fig. 15(A)). The correlationcoefficient between the two datasets was lowest (0.65) inJogeva (Fig. 15(B)). According to the accommodationstatistics, there are very few overnight tourists in JogevaCounty during the winter months. Despite the low numberof overnight stays, there is a relatively high number oftourists in Jogeva County, even in the winter months,which may be caused by a high transit flow of tourists viaJogeva County.Comparison of two datasets is a very clear way to

compare spatial supply and demand in tourism. Thisinformation has been used in Estonia to set regionalobjectives for tourism investment programmes. Two

ommodation and call activities by counties.

n Laane (A) county (r ¼ 0.98) and Jogeva (B) county (r ¼ 0.65).

ARTICLE IN PRESSR. Ahas et al. / Tourism Management 29 (2008) 469–486484

features characterise counties with low correlations such asJogeva, Valga, Voru, and Rapla. Firstly, they are lesstourism oriented, without a developed tourism infrastruc-ture such as accommodation establishments. The secondaspect is that the importance of tourist transit flowcharacterises those counties that have a low correlation.Jogeva, Voru, Valga, and Rapla counties are in the area ofinternational transit traffic, and many tourists drivethrough these counties. The low correlation betweenaccommodation and mobile positioning statistics in someareas actually shows high development potential: highdemand and low supply in the tourism industry. Mobilepositioning data offers new methods and approaches inmarketing analysis and planning, which will be elaboratedbelow.

5. Discussion

5.1. Access to data

In conclusion, it is important to emphasise the mostimportant aspects of using passive mobile positioning data.In our opinion these case studies supported our under-standing that passive mobile positioning data is a promis-ing source for tourism research and management. Thedatabase is spatially and temporally more precise thanregular tourism statistics, and thus this may open some newperspectives. Today, the main bottleneck is access to data.There are several reasons why mobile operators do notwish to share data with scientists and LBS developers.There is the risk of losing public trust and subscribersbecause of the unclear use of positioning data or because ofa single mistake with data security. The surveillance issue isimportant and closely tied to the question of access tooperators. There are a number of issues involving privacyand surveillance that need to be discussed before operatorsopen databases. The operators are also not interested insharing data because today research projects and applica-tions are not big enough to attract funding with a businessplan and possible revenues. Therefore there is a need toinvolve operators through other aspects such as publicityfor participation in R&D projects or to advance innova-tion. It is not only publicity that attracts mobile operators,but innovative operators are open to seeking alternativesources of income sources from expensive infrastructurethey have to develop for a major product—call minutes.

5.2. Developing the method, applications and products

One important issue with using passive mobile position-ing data is sampling; mobile phones are widespread, butvery few studies have been performed about the everydayuse of phones. Who uses the phone, when and for whichpurposes is an important question. Phone use in differenttourist segments is also significant. The quality ofpositioning data was earlier compared and discussed herein connection with accommodation statistics. The correla-

tions between the two databases were high (up to 0.99) indeveloped tourism regions such as Western Estonia. Inareas with lower correlations (the lowest being 0.65 inJogeva County), infrastructure is less developed, andtransit tourism via highway and railway is of highimportance. Such areas may interest investors as placeswith a great demand for services.We introduced some applications of our analyses, mostly

in the consulting of public authorities in Estonia concern-ing regional planning and spatial planning. The first groupof applications, statistical overviews and simple descrip-tions of tourism, are in demand at the municipal andcounty levels. The authorities use such information tocompile local development plans and funding applicationsthat are submitted to governmental or EU fundingagencies. Similar data is also needed for spatial planningand strategic planning projects. The second type ofapplications is related to more specific and more detailanalyses, such as trend analyses or space–time variabilitystudies. Tourism authorities and some municipalities askedus more detail questions about trends in the presence ofsome nationalities in certain regions or about the space–-time impacts of single-tourist attractions such as festivalsfunded by the authorities. The evaluation of tourism inless-habited areas such as nature reserves was alsosuccessful, as traditional questionnaires are too expensiveto be used to reach isolated visitors. In particular, themobile-based data may be critically important for theformulation of development plans for rural destinationsthat lack marketing resources and need better networkingto attract foreign tourists. The positioning data couldhighlight some inherent deficiencies in the quality andsafety of tourism marketing, principally between thepromotion of tourism sites and the sustainability of suchsites in the wake of mass tourism. The knowledge of thetime and space of the routine routes used most frequentlycan be exploited to mark periods of critical capacity, inorder to avoid negative social, environmental or culturalimpacts. Also, the underutilisation of tourism resourcescould be identified in order to shift visitors’ flows toalternative times, paths and places. There are some otherapplications and perspectives with public services con-nected with the development of real-time monitoringsystems and logistical systems in the destination country.Unfortunately, we do not have experience, and therefore

will not introduce in depth the issues that arise from LBSbusiness applications. Passive mobile positioning data haspotential to be applied for the development of businessesfor the better management of tourism demand and supplyand intermediate services. Tourists need more informationto have better services and to be satisfied with destinations.Location-aware information composed personally anddelivered to a mobile phone is a field in which many LBSactivities have taken place in recent years around theworld. There are now mobile travel guides, location-awareinformation phones, mobile travel guides with mixedreality and ubiquitous solutions. Nevertheless, very few

ARTICLE IN PRESSR. Ahas et al. / Tourism Management 29 (2008) 469–486 485

of them have achieved success outside the testing phase.From the supply side, service providers wish to reachpotential clients who enter an area. This is not only amatter of advertisement, but also an issue of targeting:reaching the right person for a certain product. This is alsoa question of forecasting demand, and using such a forecastto offer better services. If hotels, restaurants and logisticscan estimate or see the actual number/time of touristsentering an area, they can save costs, offer better quality orsustain the environment.

Nevertheless, it appears that passive mobile positioninghas greatest potential today with intermediate businessesand public administration tasks. These bodies have thecapacity to use information about potential tourist flows,timing and locations to find the best solutions for thesatisfaction of supply and demand and to provideimportant guarantees for both ends. New data sourcesand methods such as real-time implications with position-ing data can lead to new developments in academic tourismgeography, since there have been pessimistic notes aboutthe future of academic studies (Hall & Page, 2006). If LBSbusinesses really begin to take off, there will at least be aneed for geographical knowledge and scientific methodsduring product development. This may open morepossibilities for academic bodies.

5.3. Method for future

There has been much scepticism about the potential ofmobile positioning in geography. Most criticisms areconnected with questions about the need for anotherquantitative dataset, the spatial accuracy of positioningdata and privacy/surveillance issues, which tend to becomeincreasingly important with all ICT developments. Thesecriticisms are understandable in a world of the ‘‘war withterror’’ and the failure of the first round of LBSdevelopments. Nevertheless, the authors of this paperbelieve that together with rising mobility and changes insociety and the environment, there is a growing need tomanage resources better. As almost everyone has mobilephones, the digital spatial track recorded by positioningcould help organise life better. Therefore we have toremember that the most important aspect of testing mobilepositioning data and methods today is that we aredeveloping this method for the future. The world of ICTand society is developing so fast that today’s problems andlimitations will easily be surpassed tomorrow or the dayafter tomorrow. This also pertains to the topic of datacollection and cooperation with mobile operators and thestandardisation of data and analysis methods until theyreach the level of official statistics. The most promisingarea of development may be real-time monitoring systemsand live maps using local and international visitors’statistics from operators’ networks (Joint Space, 2007).Nevertheless, the most unpredictable issue is related to thefear of privacy and surveillance. This unpredictable aspectmay change the development of the entire LBS sector. Due

to the fear of surveillance, however, the data can also havemore value for people participating in the studies and forpeople using it as participatory tool. This ‘‘imaginary’’value of ‘‘my presence somewhere’’ can develop mobilepositioning systems to become powerful participatorytools.Passive mobile positioning is only one type of data for

the investigation of tourism and geography that can beobtained from mobile positioning. The use of active mobilepositioning or active tracing experiments is a rapidlydeveloping direction in ICT (Ahas et al., 2006). This isalso used in studying tourism behaviour in destinationcountries. Active tracing has its advantages, because it ispossible to model the sample, and question or interviewparticipants. This allows one to link the quantitativemovement track with qualitative data about tourists’opinions and values–the SPM (Ahas & Mark, 2005).Active mobile positioning is spatially more accurate thanpassive positioning data. Because of the success of A-GPStelephones, and the rapid development of EuropeanGalileo, active mobile positioning will soon have greatspatial accuracy, i.e. to within a few metres.

Acknowledgements

This article was supported by Target Funding ProjectNo. SF0180052s07 and Grant of Estonian Science Foun-dation No. ETF7204.

References

Ahas, R., Aasa, A., Mark, U., Pae, T., & Kull, A. (2007). Seasonal

tourism spaces in Estonia: Case study with mobile positioning data.

Tourism Management, 28(3), 898–910.

Ahas, R., Aasa, A., Silm, S., & Tiru, M. (2007). Mobile positioning data in

tourism studies and monitoring: Case study in Tartu, Estonia. In M.

Sigala, L. Mich, & J. Murphy (Eds.), Springer computer science:

Information and communication technologies in tourism (pp. 119–128).

Ahas, R., & Laineste, J. (2006). Technical and methodological aspects of

using mobile positioning in geographical studies. In K. Pae, R. Ahas,

& U. Mark (Eds.), Joint space. Open source on mobile positioning and

urban studies (pp. 37–43). Tallinn: Positium.

Ahas, R., Laineste, J., Aasa, A., & Mark, U. (2007). The spatial accuracy

of mobile positioning: Some experiences with geographical studies in

Estonia. In G. Gartner, W. Cartwright, & M. P. Peterson (Eds.),

Location based services and telecartography. Berlin: Springer Lecture

Notes in Geoinformation and Cartography.

Ahas, R., & Mark, U. (2005). Location based services—New challenges

for planning and public administration? Futures, 37(6), 547–561.

Beeho, A. J., & Prentice, R. C. (1995). Evaluating the experiences and

benefits gained by tourists visiting a socio-industrial heritage museum:

An application of ASEB grid analysis to Blists Hill Open-Air Museum,

The Ironbridge Gorge Museum, United Kingdom. Museum Manage-

ment and Curatorship, 14(3), 229–251.

Buhalis, D., & Licata, M. C. (2002). The future eTourism intermediaries.

Tourism Management, 23(3), 207–220.

Casino, V. J. D., & Hanna, S. P. (2000). Representations and identities in

tourism map spaces. Progress in Human Geography, 24(1), 23–46.

DIRECTIVE, 2002/58/EC: Directive on privacy and electronic commu-

nications of the European Parliament and of the Council of 12 July

2002 concerning the processing of personal data and the protection of

ARTICLE IN PRESSR. Ahas et al. / Tourism Management 29 (2008) 469–486486

privacy in the electronic communications sector. /http://europa.eu.int/eur-lex/pri/en/oj/dat/2002/l_201/l_20120020731en00370047.pdfS.

ESA. (2004). Tourism 2003. Tourism in 2003. Ministry of Economic

Affairs and Communications. /http://www.mkm.ee/failid/0504_

Turism_2003.pdfS.

Eurobarometer. (2006). Special Eurobarometer 269/Wave 66.1–TNS

Opinion & Social. 2006.

Fennell, D. (1996). A tourist space–time budget in the Shetland Islands.

Annals of Tourism Research, 23, 811–829.

Frihida, A., Marceau, D. J., & Theriault, M. (2004). Development of a

temporal extension to query travel behaviour time paths using an

object-oriented GIS. GeoInformatica, 8(3), 211–235.

Graefe, A. R., & Vaske, J. J. (1987). A framework for managing

quality in the tourist experience. Annals of Tourism Research, 14,

390–404.

Hall, C. M., & Page, S. (2006). The geography of tourism and recreation.

Environment place and space. Routledge.

Haywood, K. M., & Muller, T. E. (1988). The urban tourist experience:

Evaluation satisfaction. Hospitality Education and Research Journal,

18(2), 453–459.

Hagerstrand, T. (1968). Innovation diffusion as spatial process. Chicago:

University of Chicago Press.

Kwan, M. P. (2000). Interactive geovisualization of activity-travel patterns

using three-dimensional geographical information systems: Methodo-

logical exploration with a large data set. Transportation Research C, 8,

185–203.

Lew, A., & McKercher, B. (2006). Modeling tourist movements.

A local destination analysis. Annals of Tourism Research, 33(2),

403–423.

Ratti, C., Frenchman, D., Pulselli, R. M., et al. (2006). Mobile landscapes:

Using location data from cell phones for urban analysis. Environment

and Planning B, 33(5), 727–748.

Spinney, J. E. (2003). Mobile positioning and LBS applications.

Geography, 88, 256–265.

Timmermans, H., Arentze, T., & Joh, C.-H. (2002). Analysing space–time

behaviour: New approaches to old problems. Progress in Human

Geography, 26(2), 175–190.

Unwin, T. (1998). Tourist development in Estonia: Images, sustainability,

and integrated rural development. Tourism Management, 4(4),

265–276.

Wang, D., Cheng, T. (2001). A spatio-temporal data model for activity-

based transport.

Worthington, B. (2003). Change in an Estonian resort: Contrasting

development contexts. Annals of Tourism Research, 30(2), 369–385.

Internet sources

Positium. (2007). Positium homepage. /www.positium.comS; Last visited

23.01.2007.

EMT. (2007). EMT homepage. /www.emt.eeS; Last visited 10.01.2007.

ESA. (2007). Estonian Statistical Board homepage. /http://www.stat.eeS;

Accommodation statistics. Last visited on 23.01.2007.

EAI. (2007). Estonian Data Protection Inspectorate. /http://www.dp.go-

v.eeS; Last visited 23.01.2001.

Joint Space. (2007). Joint Space–Online Positioning Enviropnment.

/www.jointspace.eeS; Last visited 12.06.2007.