Full Terms & Conditions of access and use can be found at
http://www.tandfonline.com/action/journalInformation?journalCode=cjut20
Download by: [Masarykova Univerzita v Brne] Date: 16 November 2017, At: 05:52
Journal of Urban Technology
ISSN: 1063-0732 (Print) 1466-1853 (Online) Journal homepage: http://www.tandfonline.com/loi/cjut20
Using Mobile Positioning Data to Model Locations
Meaningful to Users of Mobile Phones
Rein Ahas , Siiri Silm , Olle Järv , Erki Saluveer & Margus Tiru
To cite this article: Rein Ahas , Siiri Silm , Olle Järv , Erki Saluveer & Margus Tiru (2010) Using
Mobile Positioning Data to Model Locations Meaningful to Users of Mobile Phones, Journal of
Urban Technology, 17:1, 3-27, DOI: 10.1080/10630731003597306
To link to this article: http://dx.doi.org/10.1080/10630731003597306
Published online: 30 Mar 2010.
Submit your article to this journal
Article views: 1071
View related articles
Citing articles: 84 View citing articles
Using Mobile Positioning Data to Model Locations
Meaningful to Users of Mobile Phones
Rein Ahas, Siiri Silm, Olle Ja¨rv, Erki Saluveer, and Margus Tiru
ABSTRACT The article introduces a model for the location of meaningful places for mobile
telephone users, such as home and work anchor points, using passive mobile positioning
data. Passive mobile positioning data is secondary data concerning the location of call
activities or handovers in network cells that is automatically stored in the memory of
service providers. This data source offers good potential for the monitoring of the
geography and mobility of the population, since mobile phones are widespread, and
similar standardized data can be used around the globe. We developed the model and
tested it with 12 months’ data collected by EMT, Estonia’s largest mobile service provider,
covering more than 0.5 million anonymous respondents. Modeling results were compared
with population register data; this revealed that the developed model described the
geography of the population relatively well, and can hence be used in geographical and
urban studies. This approach also has potential for the development of location-based
services such as targeting services or geographical infrastructure.
Introduction
Population mobility is growing rapidly in the contemporary world. According to
the mobility paradigm (Sheller and Urry, 2006), movement has become a phenomenon
in its own right in the twenty-first century; employment is more and more
mobile, tourism has become a lifestyle, and it is common to own multiple residences
and jobs. Rising everyday mobility has been combined with the rapid
growth in the use of mobile phones and other mobile communication devices.
In order to assess and analyze the location of individuals and populations in
this mobile world, new methods and approaches are needed. Traditional census
and population registers are solid sources for long-term processes. For the short
term and everyday mobility, more flexible methods such as various registers
and indirect databases (Raymer et al., 2007), satellite-based methods (Chen et al.,
2006), or modern sensing technologies (Kwan, 2000; Eagle and Pentland, 2005;
Shoval, 2007) are needed.
One of the proposed methods for developing such a monitoring tool is mobile
positioning or mobile telephone tracking (Mountain and Raper, 2001; Spinney,
2003), also called the social positioning method (Ahas and Mark, 2005) or cellular
census (Reades et al., 2007). Mobile positioning is often considered to be a novel
and exciting source of information for investigating the spatial dynamics of
human society, while at the same time the number of published studies is small
Journal of Urban Technology, Vol. 17, No. 1, April 2010, 3–27
1063-0732 Print/1466-1853 Online.
Copyright # 2010 by The Society of Urban Technology
DOI: 10.1080/10630731003597306
All rights of reproduction in any form reserved.
Downloadedby[MasarykovaUniverzitavBrne]at05:5216November2017
because of problems concerning limited access to such data and privacy issues. A
number of interesting works regarding urban studies (Ratti, 2005; Reades et al.,
2007; Ahas et al., 2007a), tourism studies (Ahas et al., 2008a), transportation
studies (Asakaura and Hato, 2004; Herman, 2006), and GIS (Nurmi and Koolwaaij,
2006; Gartner, 2004) however, have been published in recent years.
There are many methods and approaches that can be used to locate mobile
telephones. Technical solutions vary from handset-based systems with special telephone
software to satellite navigation and peer-to-peer positioning tools using
Bluetooth. The most positive aspect of mobile positioning is that nowadays
mobile phones are becoming pervasive in developed and developing countries
alike. In order to conduct population and mobility studies based on positioning
data, a great deal of preparatory work is necessary, because data needs to be processed
and assessed, and the appropriate methods must be developed.
The objective of this paper is to develop a model for locating places that are
meaningful to mobile phone users. This is done by using passive mobile positioning
data. Meaningful places or meaningful locations (Nurmi and Koolwaij, 2006)
are defined as regularly visited places that have meaning for individuals. Technically,
they are similar to personal anchor points; home and work anchors are the
most common among them. With the rising popularity of the mobile lifestyle and
mobile communication, there is an increasing number of regularly visited places,
and mobile phones can be used to detect those places. Passive mobile positioning
data are a secondary source of geographical data that are automatically stored in
the memory files and logs of mobile operators (Ahas et al., 2008b). Here we introduce
our model for determining the geographical location of meaningful places
using an Estonian example. The modeled results of the research were compared
with the data of the Estonian Population Register.
This quantitative-based model is an important step in developing alternative
tools to monitor the everyday mobility of the population using various tracking
data and to develop location based services (LBS) for mobile phone users.
Mobile positioning data are becoming more and more common in geographical
studies, and LBS are finding new markets and products around the globe. Therefore,
there is growing interest in such “mobile geography” as a current model for
anchors, not only from a geographical perspective but also from the information
technology (IT) sector for the personalization of mobile services. The issues of
privacy and surveillance are important aspects of mobile positioning. The focus
of this paper does not allow us to discuss this in depth, but we cover some
topics related to passive positioning in our discussion of the data.
Theoretical Framework
Passive Mobile Positioning
Mobile positioning means tracing the location coordinates of mobile phones.
There are different frameworks for positioning, for instance handset-based,
network-based or GPS-based. In order to locate phones, radio waves are used,
and positioning is done using various methods, such as cell ID, triangulation
with direction angle, and/or distance from an antenna. Different positioning
methods are used because of the different network standards (GSM, CDMA,
3G), and there are different purposes for positioning (Zhao, 2002). In addition,
4 Journal of Urban Technology
Downloadedby[MasarykovaUniverzitavBrne]at05:5216November2017
the use of mobile positioning data in geographical studies has different
approaches and algorithms (Roos et al., 2002).
Mobile positioning can be divided into passive and active positioning. Active
mobile positioning is used for mobile tracking in which the location of the mobile
phone is determined (asked) with a special query using a radio wave (Ahas et al.,
2007a). Passive mobile positioning is data which is automatically stored in
memory or log files (billing memory; hand-over between network cells, Home
Location Register, etc.) of mobile operators (Ahas et al., 2008a). The easiest
method for passive mobile positioning is “a billing log” that is recorded for
called activities. Any active use of a mobile phone (call and SMS messages in
and out, GPRS, etc.) is deemed to be call activity. Passive mobile positioning
data is normally collected to the precision of network cells. Every cell has a
certain geographical coverage area and unique identity code, and, therefore,
this method is called Cell ID. Mobile operators can aggregate anonymous geographical
data from log files, such as location points or movement vectors, and
researchers can use these in surveys for scientific purposes. Issues of privacy
and surveillance are very important aspects of any mobile positioning data.
Other sources than call activity are also used for passive positioning, such as
Erlang of antennae (Reades et al., 2007).
When using passive mobile positioning data, as we do in the current paper, it
is important to explain some basic definitions of the topic. The cellular network is
based on a set of base stations, which usually have one tower and several directed
antennae. The radio coverage of a single antenna forms a network cell; several
antennae form a cellular network. Every network cell in a mobile phone
network has a unique ID and geographical coordinates, and the location of a
phone in the cell can be easily determined for every call activity.
The size of a network cell and all cellular networks is not fixed; the phone normally
switches to the closest antenna or the one with the strongest radio coverage
or best visibility. If the network is crowded or visibility is disturbed, the phones
can be switched not to the nearest station but to any other in the neighborhood.
The maximum distance from a handset to an antenna in the GSM network is
less than 35 km. There are amplified antennae used in GSM networks in less
inhabited or coastal areas that cover greater distances.
Conceptual Framework for Using Mobile Positioning Data
Our objective is to develop a model that recognizes meaningful places visited by
single telephones (persons). These places can also be referred to as regularly
visited places of personal anchor points such as home and work, etc. It has been
noted, however, that with the rising mobility of individuals, the dominance of
home and work anchors is decreasing, and people also spend their time in other
important locations. Therefore, the term “meaningful places,” which is very relevant
to mobile positioning-based data, is often used. To use mobile positioning
data to model locations of regularly visited places, we must conceptualize the interactions
between mobile phones and space: Is the location of a mobile call something
special, or it is just another location of another activity? Several researchers
have been discussing the location of mobile phones and mobile calls. Most of
them have pointed out that a wireless phone is a special place with its own
spatial features. Finnish researcher Timo Kopomaa called a mobile phone, “an
Using Mobile Positioning Data to Model Locations Meaningful to Users 5
Downloadedby[MasarykovaUniverzitavBrne]at05:5216November2017
instrument for maintaining contact: the mobile phone can be viewed as a place”
(Kopomaa, 2004, p. 268). He also labeled the mobile phone as a “third place”
meaning the third important location after home and work. The third place was
defined by Oldenburg (1989) as a place of social interaction with communication
networks. The locations of home and work places in mobile networks may be
different from their actual geophysical locations. But they are recognizable as
regularly visited places in the third place, the mobile telecommunications network.
The conceptualizing of mobile positioning data begins with the need to
remember that mobile calls are social events; every “call out” or “call in” involves
another person or call partner with a telephone. Therefore, we must remember
that each call also depends on the availability of a communication partner. This
is why, even if some people wake up early or drive through the night they do
not call as often, as their potential call partners are asleep. Indeed, the calls of
some are calls with other time zones.
There are many interesting methods for investigating the geographical and
social aspects of networks of personal communication. In this study we focus
on the location of persons calling someone by phone. In connection with mobile
telephones, we must also keep in mind that people either do or do not use
phones in certain places and for certain activities. The culture of phone use has
not been systematically studied in Estonia, but many countries have learned
about the use of phones. For example, automobiles or public transportation are
among the most popular places to make calls, and for the same reason, mobile
phones are a “modern” source of traffic accidents. As we study the location of
calls, it is important to know that people may call during or after certain events
(the New Year, after a football match has been won) or at times when emotions
are stronger than normal, and those times and locations may not leave a proportional
trace in the data.
Because of privacy issues, the massive mobile positioning datasets used in
several countries (Ahas et al., 2007a; Reades et al., 2007) do not have many identifications
or personal features of the owners of the phones studied. Normally those
databases have huge quantities of locations or “dots”. Those dots can serve as
sources for the modeling of meaningful places for people.
Approaches for Determining Meaningful Places
There are many conceptual approaches to the description and analysis of people’s
everyday movements. The Swedish geographer T. Ha¨gerstrand (1970) conceptualized
geographical movement by including a time dimension in the data model. He
defined paths as movement corridors and stations as visited places which are
typical for most of humans. Space-time is a dimension we can also use in the conceptualization
of the given mobile positioning data, since both chronological and
chorological dimensions are exact and vast in the database. The lack of additional
personal information and identifications forces us to disregard models and
approaches that rely on the individual potentials of travel as mental maps,
action space, and perceptual space using this positioning data (Scho¨nfelder and
Axhausen, 2004).
The anonymous passive mobile positioning database we use is more applicable
with interpretations based on the actual activity spaces concept that
describes the structures of realized locational choices for single travelers and is
6 Journal of Urban Technology
Downloadedby[MasarykovaUniverzitavBrne]at05:5216November2017
also referred to as micro-geographical activity space (Scho¨nfelder and Axhausen,
2004). This approach has been used by different schools for many decades
(Du¨rckheim, 1932). Activity spaces represent the distribution of places visited
and the space that contains those places frequented over a period of time. Activity
spaces are geometric indicators of observed or realized daily travel patterns
(Axhausen et al., 2002; Scho¨nfelder and Axhausen, 2004; Dijst, 1999). The study
of the geometry of activity spaces may be a possible solution for the handling
of our passive mobile positioning data. Some examples of possibilities are the
theoretical concepts of confidence ellipses (Scho¨nfelder and Axhausen, 2004;
Schwarze and Scho¨nfelder, 2001). The confidence ellipse method using the Visar
program was successfully used in a study of gender differences in mobile
positioning data (Silm et al., 2008). Other GIS-based analytical tools have also
been proposed, such as kernel densities or geometrical analyses of locations.
The investigation of the geometry of activity spaces is insufficient if there are
not any additional data available about individuals and their activities.
This is why the further use of passive positioning data depends on the
implementation of the anchor points concept. Anchor points are locations
where people regularly stay (Golledge and Stimson, 1997). Besides anchor
points, the terms bases (Mitchell and Rapkin, 1954; Kutter, 1973; Vidakovic cit
Dijst, 1999) or core stops (Schank and Abelson, 1977) have also been used.
Anchor points can be divided into two meanings. “Common anchors” are significant
places in the environment that are commonly recognized and used as key
components of cognitive maps (Dijst, 1999). “Personalized anchors” are related
to a person’s activities (e.g., a specific work place or home-base) (Golledge,
1990). The meaning of anchor points is described from different perspectives by
different authors. Our interest here is in the GIS-based analysis of huge datasets.
For this purpose, there is one relevant concept—“a personal network of usual
places,” consisting of activity places and the routes between them (Flamm,
2004; Flamm and Kaufmann, 2006). Flamm and Kaufmann determine a certain
number of daily life centers among the activity places, i.e., places where
individuals usually spend a considerable amount of time and which they consider
important in the conduct of their everyday lives; these are typically the home
and the workplace. It can be assumed that these daily life centers represent
“territorial anchor points” of the personal activity space, and that they are most
closely interconnected to other activity places by travel routes. Near daily life
centers are clusters of activity places such as grocery stores for daily shopping,
automatic teller machines, restaurants, etc. One important issue was to define
the parameters used to identify centers of daily life and to differentiate usual
activity places from others (Flamm and Kaufmann, 2006). Flamm and Kaufmann
used the Moby drive dataset and determined anchor points on the basis of the
following criteria: a) Centers of daily life were defined as places where an
individual spent at least 67 hours over the entire six-week survey period. This
value was chosen in order to include all places where individuals were employed
for 20 percent of the time. Usual places were defined as activity places that were
visited at least three times during the survey period, i.e. places with at least a
bimonthly visit frequency. This criterion appeared to be most appropriate, since
the survey period of six weeks did not allow detection of a monthly visit
frequency (Flamm and Kaufmann, 2006).
From this theoretical and methodological basis, we began the compilation
of our model or framework for the determination of meaningful places (same as
Using Mobile Positioning Data to Model Locations Meaningful to Users 7
Downloadedby[MasarykovaUniverzitavBrne]at05:5216November2017
personal anchor points or regularly visited places here) using passive mobile
positioning data. As mentioned above, passive mobile positioning data have
their peculiarities as there are a huge number of location points over a relatively
long time period, and very little information about respondents’ social features.
Compatible methods have been developed for such data sources as the number
of GIS-based analyses of mobile trajectories (Mountain, 2005; Zhou et al., 2007)
and GPS-based experiments in travel behavior studies (Wolf et al., 2001).
Computer scientists have also performed interesting studies to determine
meaningful places from mobile positioning data and to develop algorithms for
location-based services with different methods such as clustering movements
and places (Kang et al., 2004). It must be mentioned that the clustering method
developed by Nurmi and Koolwaaij (2006) and Nurmi and Bhattacharya (2008)
is another approach for determining important locations and movement types
in cellular networks.
Data
Passive Positioning Data
In the current study, we use data from the databases of the largest Estonian mobile
operator—EMT. EMT covers nearly 99.9 percent of the total land area of Estonia,
which measures 46,000 km2
. Eurobarameter estimates that 87 percent of the
Estonian population of 1.38 million have mobile phones (Eurobarometer 2007).
The database we use in the current study is passive positioning data—the
locations of all “calls out” (calls initiated by the respondent, not received calls)
with the precision of Cell IDs made in the Estonian largest network, EMT,
during a period of one year. EMTuses network hardware and positioning middleware
from Ericsson Ltd (Ericsson, 2008). The system uses Ericsson’s GSM, GPRS/
EDGE and WCDMA/3G network technologies and latest mobile positioning platform,
MPS 9.0. The entries include the following parameters for every outgoing
call of a phone registered in the network: a) the exact time of the call activity; b)
the random ID number for the phone (not related to phone or SIM card
number); c) the cell ID with the geographical coordinates of the antenna. Due to
privacy issues, we do not possess any personal information about the respondents,
but only a randomly assigned ID for every phone. The random ID generated
by the operator enables us to identify the calls made by one person during
the study period. There are hundreds of locations of calls for every ID, and we
begin our analysis with this data. An example of the database is: Time September
8. 2007. 22:03:11; ID 64353; Location E27-44-39.00 N59-25-49.00.
Data are gathered by the company Positium LBS, which has a contract with
the two major mobile operators in Estonia regarding the use of LBS data (Positium
2008). The current model was developed through a cooperative arrangement of
the Positium LBS (authors E. Saluveer and O. Ja¨rv–) and the chair of human
geography at the University of Tartu. The modeled data we present herein were
initially used to analyze traffic on the Tallinn-Tartu Highway (Ja¨rv et al., 2007)
and to analyze urban sprawl (Silm et al., 2007; Silm and Ja¨rv, 2007; Ahas et al.,
2007b). The Positium LBS database is powered by the open-source PostgreSQL
database manager. As a monthly average, the database contains the calls of
592,553 people for every 12 months from November 1, 2006 to October 31, 2007.
The average number of calls per month was 65.3 million. There are a total of
8 Journal of Urban Technology
Downloadedby[MasarykovaUniverzitavBrne]at05:5216November2017
783 million location points in the database for the researched period. The distribution
according to the ID and the number of calls is provided in Figure 1.
There is an average of 100-120 calls per ID per month.
Privacy Issues
The use of positioning data brings up the issue of privacy and surveillance. This is
the major concern of phone holders, operators, researchers, and the general public.
Therefore, the research team together with the mobile operator and the Estonian
Data Protection Inspectorate checked carefully the accordance of data use with
Estonian legislation and EU directives (European Parliament, 2002). Our study
and discussions concluded that the personal privacy of respondents is protected.
There is no personal information connected with movement vectors, and the
generalization level of the analysis does not allow the identification of persons
on geographical or temporal grounds. It is not possible to extract individual
movements from the data. Nevertheless, there is concern about issues of
privacy and ethics, as any use of mobile positioning data is very sensitive in
this respect. Our approach in Estonia was to make all research and results
transparent, and we published our plans and results step by step in newspapers,
on TV, and also on a website during the eight years of our studies in this field.
Thus, everyone could see how “their” data were used and what the benefit of
the research was. Nevertheless, we received many bad comments as feedback,
but the mobile operator was happy with the transparency and the public
comments, and we got the “green light” for the study.
In the current research, the data of the Population Register from January 1,
2007 at the municipality level were used as comparative material alongside the
passive mobile positioning data.
Geographical Distribution of the Mobile Network
The mobile network is usually distributed unevenly over a territory, following
population density patterns and transportation infrastructure. As a result, the
outcome in a geographical sense is fairly objective, since positioning accuracy is
Figure 1. The distribution of the number of IDs and Calls in the passive positioning database used
per month
Using Mobile Positioning Data to Model Locations Meaningful to Users 9
Downloadedby[MasarykovaUniverzitavBrne]at05:5216November2017
smaller in less populated areas than in densely populated areas, and the overall
error is smaller. Figure 2 represents the EMT mobile network in Estonia as of
January 1, 2007. The mobile network is in a constant state of flux, with new
base stations and antennae developed by all progressive operators.
In the present survey, local municipalities have been used as the subjects of
research, and the data from mobile antennae have been generalized to match
the level of local governmental units. In reality, mobile networks do not follow
the borders of local governmental units. As a result, the data from mobile antennas
are restricted to the local municipality on whose territory the mobile antenna are
located. As a result, there are 12 local municipalities in Estonia (mostly in rural
areas) about which there are no data, because there are no mobile antennas situated
in those areas. For these units, a mean value has been calculated, based on
the average data from other neighboring governmental units.
Description of Model
General Points
The passive positioning database is composed of the location and time of outgoing
calls and of the random ID assigned to the respondent, which is not connected to
the phone number but remains constant for every phone in the network. Based on
random ID, the person’s calls can be linked to each other through the study
period. Personal anchor points, presumably home, work and other places, can
be found as the locations the respondent visits regularly (hereafter, anchor
points). Anchor points are important variables in describing humans’ behavior
Figure 2. The EMT mobile network (mobile antennas and tiessen’s polygons), counties, and major
towns
10 Journal of Urban Technology
Downloadedby[MasarykovaUniverzitavBrne]at05:5216November2017
in time and space. Anchor point modeling is one of the possibilities for making
useable the anonymous data of passive mobile positioning, GPS tracking, etc.
In accordance with the frequency and regularity of the respondents’ calls, we
have chosen a calendar month to be the timestep for determining anchor points in
the current paper. Thus we analyzed 12 months separately for every person, and
the results are presented here as an average of 12 analyzed months for every
person. Presumably, periods of one week would also produce results, but we
believe that in Estonian conditions, the outgoing calls database yields better
results for one-month periods.
The accuracy of pinpointing anchor points is at the level of the service area of
a mobile antenna (a network cell) (See Figure 3). In different network systems, it is
actually possible to divide the network cell into smaller cells, but this depends
on the methods of data collection and storing. In this survey, we have used the
Ericsson system, which dispenses sites of mobile antennas as a network cell
because of the optimal price and quality of the data.
Terms Used
Respondent—a person (sometimes the term phone is also used), who owns a cell
phone connected to the EMT network, who has performed calls on it, and who
has been added into the passive positioning database by their random ID.
Regular Cells—network cell regularly visited by one respondent and from which
the respondent has made calls on at least two different days a month.
Random Cell—a network cell from which the one respondent has made calls on
only one day a month.
Meaningful Place—a personal anchor point or regularly visited place which has a
significant meaning in the everyday life of the respondent.
Everyday Anchor Point—anchor point in which the respondent has spent time on
most days, and which have thus been assigned as home or work places.
Secondary Anchor Point—anchor points that have lower visiting regularity than
everyday anchor points.
Home Anchor Point—An everyday anchor point, at which the probable location of
the respondent’s home is determined, based on the model.
Work-Time Anchor Point—an everyday anchor point, at which the probable worktime
location of respondents is determined, based on the model. The anchor is
called a work-time location because it is not possible to differentiate between
work, school, and other activities in the place where a person regularly and
most often spends time in business hours during a month.
Multifunctional Anchor Point—an everyday anchor point in which the home and
work-time anchor points are located in the same network cell and cannot be
separately identified.
Description of Model
A model for detecting meaningful places or personal anchors of persons consists
of eight steps (See Figure 3), of which each stage actually contains a complex
PostgreSQL database manager query or framework. The important stages of
model calculation have been brought out in the following list.
Using Mobile Positioning Data to Model Locations Meaningful to Users 11
Downloadedby[MasarykovaUniverzitavBrne]at05:5216November2017
In order to develop the model, we studied in detail 14 selected friends or relatives
(five urban, five rural, and four suburban residents) whose logs of calls out
were recorded during 12 months. We asked personally about all locations of calls
from all respondents during interviews. The model was developed step by step,
Figure 3. Anchor point determining model
12 Journal of Urban Technology
Downloadedby[MasarykovaUniverzitavBrne]at05:5216November2017
testing entries in log file and meaning of locations of those calls during 12 months.
We compared the modeled home and work-time anchor points with each
individual’s actual places of residence and work. Figure 4 shows the location of
one of our colleague’s anchor points in the city of Tartu in November 2007.
Determining Points of Regular Cells and Separating Them from Random Cells. For
every respondent (i.e., ID), the number of calls made is determined for each
separate network cell. Based on this information, regular cells (network cells)
where the respondent has performed calls on at least two separate days a
month will be separated. Random cells from which the ID has made calls on
only one day of the month will be left out of further analysis.
Removal of IDs with Too High or Too Low a Number of Calls Made. The next stage
of the model removes respondents with too high or too low a number of calls from
the database. For that purpose, call frequencies for every ID in every network cell
are calculated. Regular cells are sorted by the number of days with calls, and if
days overlap, by the number of calls. Then the respondents (IDs) who have
made calls from their most visited network cell on fewer than seven days a
month will be removed from the database. If the number of calls made is too
low, it is not possible to calculate the anchor points.
On the basis of the same sorted list of regular cells, respondents with too
many calls are also excluded. This exclusion is made on the presumption that
the reason for there being too many calls is an organized call procedure (service
center, etc.) or a technical device using a GSM network. Respondents are
Figure 4. One of the respondent’s anchor-point locations in the city of Tartu in November 2007
Note: Determined by means of passive positioning. The respondent’s (Prof. Ahas) activities took place
only in the city of Tartu.
Using Mobile Positioning Data to Model Locations Meaningful to Users 13
Downloadedby[MasarykovaUniverzitavBrne]at05:5216November2017
considered to have too many calls if they have made more than 500 calls a month
in their two most frequently visited network cells. In comparison, the average
number of calls made per month in the two most often visited network cells by
one person is 81.
Determining Home and Work-Time Anchor Points. The two regular cells that had
the highest number of days with calls are selected for the calculation of home and
work-time anchor points, and the rest are moved directly to the last stages of the
model, where they are called secondary anchor points. For every regular cell, the
average start time of calls (average of all calls made during a day) and the standard
deviation of call beginning times are calculated. Home and work-time
anchor points are determined from regular cells based on those two variables.
This model calculates the home anchor point from the regular cells where the
average starting time of daily calls is after 17:00. If the value of the standard
deviation of the beginning times of calls performed by a respondent is greater
than 0.175, the home location is also determined for respondents whose average
beginning time is before 17:00. If the average start time of calls is before 17:00
and the value of the standard deviation is less than or equal to 0.175, then these
regular cells are assigned to be work-time anchor points.
To distinguish home anchor points from work-time anchor points, the standard
deviation of the start times of calls is applied, since people’s work periods
have different durations, and thus the time is not sufficient for the determination
of anchors. Our analyses (Ja¨rv et al., 2007) have shown that the chronological
variability (and, therefore, also the standard deviation) of calls performed at
work is lower than that of calls made from residential locations (See Figure 5).
This is probably explained by the fact that people are at home and also use
their telephone at different times—at night, at lunch time, on the weekend, and
during holidays, etc. The time period spent at work is more compact, and
outside working hours people rarely spend time at the office. Of course, this
model has weaknesses, for example if the respondent has more than one workplace
or a very mobile job, or if they work at home.
In differentiating home and work-time anchor points, it is important to note
that in addition to the common situation of having one home and one work-time
Figure 5. Distribution of home (a) and work-time (b) anchor points according to standard deviation
of calls
14 Journal of Urban Technology
Downloadedby[MasarykovaUniverzitavBrne]at05:5216November2017
anchor point, it is also possible that the respondent has two home or two worktime
anchor points among the two most visited regular cells. At this point the
model becomes branched. If the respondent has two home or two work-time
anchor points, they move on to the next stage (Stage 4). Respondents with one
home and one work-time anchor point that have been properly determined will
skip the next stage and will be analyzed further in Stage 5.
Consideration of Neighboring Relationships in the Case of Two Home or Two WorkTime
Anchor Points. If one respondent has two home or two work-time anchor
points, this is probably due to the very common problem of “switching” or
“tossing” connections between two neighboring antenna even if the person
stays in the same room. It is typical that if the phone is situated in the border
area of the network cell, the phone will be technically handed to the neighboring
base station that has fewer users or better visibility. First the model determines
whether those parallel anchor points are situated in neighboring cells or not. If
the anchor points are situated in neighboring cells, the analysis is continued at
the same stage. However, if they are not situated in neighboring cells, the two
home or two work-time anchor points move on to the next stage (Stage 5) of
the model.
In case they were located in neighboring cells, the model erases the second
most frequently visited anchor point, first selecting the one with the higher
number of days, and when the number of days is the same, on the basis of the
number of calls. The most visited anchor point will also remain. After that, the
model will use the earlier list of regular cells of the same respondent again, and
the next most frequent, i.e. the third regular cell will be employed. The model
will re-analyze it in Stage 3 and try to determine the missing home or worktime
anchor point in the previously described manner. If the third most frequently
visited regular cell differs from the previous two in terms of parameters, it is
classified as the respective anchor point, and the respondent will then have one
home and one work-time anchor point, which will be analyzed in Stage 5.
Similarly, anchor points will move on to the next stage (Stage 5) if the new
determined anchor point is of the same type as the previous one, but is located
in a non-neighboring cell. In that case, the respondent will actually have two
locations that can be characterized as home or work-time anchors. Of course,
the real meaning of those points is not recognizable by the data we possess. If
the third most frequent anchor point does not qualify as the required missing
anchor point, according to time, standard deviation, and location metrics, the
model will search for required parameters from the data of the fourth regular
cell. The iteration can be repeated n times. In our model n equals 3, meaning
the same procedure will be executed, if necessary, until the fifth regular cell.
However, if after n iterations it is still impossible to determine the missing
anchor point, the person with two home or two work-time anchor points will
move on to the next stage (Stage 5).
Assessment of the Proportion of Days Spent at an Anchor Point. If the most frequently
visited anchor point covers more than 75 percent of the days a respondent
stayed at the two most frequently visited anchor points, it is classified as the
multifunctional (home þ work-time) anchor point. In this case, the second most
frequent anchor point will not be analyzed further, and it moves on to the
eighth stage of the model, where it will be classified as a secondary anchor point.
Using Mobile Positioning Data to Model Locations Meaningful to Users 15
Downloadedby[MasarykovaUniverzitavBrne]at05:5216November2017
Likewise, multifunctional anchor points are also formed when the respondent
is left with only one anchor point, either home or work-time.
By the end of the fifth stage there are four different possible cases of respondents’
home and work-time anchor points: 1) one home and one work-time anchor
point, 2) one multifunctional anchor point, 3) two home anchor points or 4) two
work-time anchor points. The first two cases are used to determine anchor
points and then move on to the eighth stage, where they will be classified as everyday
anchor points. The respondents with two home and two work-time anchor
points will be analyzed further in Stage 6 of the model.
Determining the Missing Home or Work-Time Anchor Point by the Addition of a
Third Point. In the case of persons with only two home and two work-time
anchor points, the model will try to determine the missing anchor point by
using the list of regular cells. The next most frequently visited regular cell will
be selected from the list and assessed according to the average time and standard
deviation of calls as in Stage 3. If the next most frequent regular cell differs from
the two previous anchor points, the respondent will have either two home and one
work-time anchor point or one home and two work-time anchor points. These
anchor points are complete and they will move on to the eighth stage, where
they are classified as everyday anchor points.
If the next most frequent regular cell employed is similar to the previous ones,
the respondent will have either three home or three work-time anchor points. If
the respondent does not have more regular cells to analyze in the database,
only two home and two work-time anchor points will remain, these four possible
variants will be analyzed further in Stage 7 of the model.
Classifying an Anchor Point as the Missing Home or Work-Time Anchor Point. In
this stage the model again attempts to create both a home and a work-time anchor
point for the respondent with two similar types of anchors. For this purpose, the
standard deviation of existing anchor points is assessed once again. In the case
of two home or two work-time anchor points, the anchor point with a greater
standard deviation is classified as the home and the other as the work-time
anchor point. In the case of three home or work-time anchor points, the standard
deviations of the first two anchor points are similarly compared to the previous
case, and the anchor point with the larger standard deviation is classified as the
home and the other as the work-time anchor point.
By the end of this stage the respondent has either two home and one worktime
anchor point, one home and two work-time anchor points or one home
and one work-time anchor point. All of these anchor points move on to the last
stage, where they are classified as everyday anchor points.
Formation of Everyday and Secondary Anchor Points. Everyday anchor points are
most frequently visited, and the possible combinations for one person are: 1) one
home and one work-time anchor point, 2) a multifunctional anchor point (home
and work-time anchors in one cell), 3) two home and one work-time anchor
point, 4) one home and two work-time anchor points. The rest of the regular
cells, which are not everyday anchor points, will be classified as secondary
anchor points.
16 Journal of Urban Technology
Downloadedby[MasarykovaUniverzitavBrne]at05:5216November2017
Results
Results of the Model Calculation
The model calculations were carried out separately for each month during the
research period that extended from November 1, 2006 to October 31, 2007. The
results are given as an average of the 12 months. The following are the results
of the model calculations by stages. Important statistics are given in Table 1.
Regular Cells. 556,681 respondents have a total of 3,589,351 points (network
cells) from which they have made calls on at least two days, meaning an
average of 6.4 regular cells per respondent. Calls placed by a respondent in a
network cell on only one day a month were interpreted to be random cells and
were removed from the calculations. Nearly 6.8 million calls or a 10 percent of
the average number of calls were removed each month. The total number of
respondents who did not place two calls in any of the network cells in one
month was 35,872, and they were removed from further calculations, since in
their case none of the network cells could be determined to be regular cells.
IDs with Too Low or Too High a Number of Calls. In the event of too few calls
(most visited network cell less than seven days a month), an average of 102,688
respondents were left out every month. This makes 18 percent of all persons
with regular cells (minimum in May: 92,883; maximum in February: 108,541).
Likewise, respondents with too many calls were excluded, i.e. those who placed
over 500 calls a month in the two most visited network cells, on average 4,200
Table 1. Modeling steps for determining anchor points and the number of entries
Stage
Anchor
points
Number
of IDs Calls
Passive positioning database 592,553 65,279,606
1 Random cells, network cells in which a person
made calls during only one day
35,872 6,842,664 Removed
Regular cells, network cells in which a person
made calls on at least two different days
3,589,351 556,681 58,436,942
2 IDs with too few calls (no cells with two days
of calls)
240,709 102,668 1,476,666 Removed
Number of regular cells after removing IDs
with too few calls
3,348,642 454,012 56,960,276
IDs with too many calls (more than 500 calls
in two of the most visited network cells)
71,583 4219 4,766,030 Removed
Number of regular cells after removing IDs
with too many calls
3,277,059 449,793 52,194,246
3–7 Home anchor points 282,572
Work-time anchor points 284,859
Persons with both home and work-time
anchor points
247,952
Multifunctional anchor points 178,458
Two home and one work-time anchor point 10,777
One home and two work-time anchor
points
13,065
8 Everyday anchor points, total 745,889 449,793
Secondary anchor points, total 2,531,170
Using Mobile Positioning Data to Model Locations Meaningful to Users 17
Downloadedby[MasarykovaUniverzitavBrne]at05:5216November2017
respondents, on average 1 percent of all respondents who have regular cells.
In our assessment, the calls were made by automated devices, call centers, or
unusually active persons. In comparison, the average number of calls for the
remainder in locations of regular visits was 81 calls per month. The division of
people by their most visited location based on the number of days spent there
and the division of people with too low or too high a number of calls placed is
presented in Figure 6.
Home and Work-Time Anchor Points. The analysis of the data found that respondents
most often have two everyday anchor points: a home and a work-time anchor
point. As a result of the model calculation, every month the model was used to locate
an average of 282,572 persons’ homes as places where people regularly spend their
nights and also used their telephone. The model also located an average of 284,859
work-time anchor points a month where time is spent during the day (at work,
school etc.). The average number of people who were simultaneously assigned a
home and work-time anchor point was 247,952 over 12 months.
A relatively large number of respondents (178,458) had only one everyday
anchor point, i.e. a unified multifunctional anchor point. The existence of one
everyday anchor point is largely connected with the size of the network cells.
There are rural areas where the density of the network is low and therefore
there is quite a high probability that a person’s home and work places are geographically
in different locations but remain in the coverage area of the same
antenna. These are therefore determined to be multifunctional anchor points.
The distribution of multifunctional anchor points in Estonia shows their dominance
in less populated rural areas and a lower number in cities and surrounding
areas. The other presumable reason for the numerous occurrences of only one
anchor point is people’s high level of domesticity. As a result of the model
calculation, a certain number of people were found to have two home and two
work-time locations, but their numbers are relatively low: 10,777 with 2 home
Figure 6. The division of people based on the number of days spent in the most visited location
Note: Anchor points highlighted in dark indicate persons with a number of calls that is too high or too
low to be used in the analysis (removed from database).
18 Journal of Urban Technology
Downloadedby[MasarykovaUniverzitavBrne]at05:5216November2017
and 1 work-time anchor points and 13,065 with 1 home and 2 work-time anchor
points. (See Table 1.)
Everyday and Secondary Anchor Points. Home and work-time anchor points,
which were calculated in stages from 5 to 7, were classified as everyday anchor
points, and their monthly average was 745,889. In addition to everyday home
and work-time anchor points, model calculations also determined the secondary
anchor points (third most frequent regular cell etc), where people reside often, yet
the frequency of visiting does not parallel the parameters related to home and
workplaces. These might be locations where people shop, run errands, spend
free time, visit family and friends, etc. During the 12 months there where
2,531,170 secondary anchor points. It was not the aim of the current research to
study and find the meaning of secondary anchor points in great depth.
Geographical Distribution of Modeled Anchor Points
The geographical distribution of respondents’ homes calculated using our model
is presented in Figure 7. To determine the location of homes, home anchor points
and multifunctional anchor points were summed. Places of residence are concentrated
in larger cities and their surroundings, and more homes are also distributed
in other more densely populated areas of Estonia, for instance the industrial
North-East (Ida-Viru County) and the agricultural centers of Southern Estonia.
The compared proportion of counties’ populations in Estonia (See Table 2)
shows that more than 38 percent of all home anchor points are focused in Harju
County, followed by Tartu County with 12.7 percent, Ida-Viru County with 8.2,
Figure 7. The density of home anchor points in municipalities modeled using the passive mobile
positioning data provided by EMT
Using Mobile Positioning Data to Model Locations Meaningful to Users 19
Downloadedby[MasarykovaUniverzitavBrne]at05:5216November2017
and Pa¨rnu County with 6.5 percent. Tallinn, the capital of Estonia, hosts 68 percent
of home anchor points in Harju County.
The geographical distribution of the modeled work-time anchor points (See
Figure 8) also matches the major patterns of population densities. Work anchor
points are composed by both work-time anchor points and multifunctional
anchor points. The comparison of the geographical distribution of home anchor
points and work-time anchor points showed that in larger cities there are more
jobs than homes. For example, in our modeled data from Tallinn, there are 94,566
homes and 108,579 work-time locations, whereas in Tartu there are 18,190 homes
and 32,348 jobs. Places of residence dominate over work-time locations in local
governmental units in the surroundings of bigger cities. The proportion of worktime
locations in Estonian counties is also quite similar to the above-mentioned
locations of home anchor points.
Comparison of the Modeled Data with the Data from the Population Register
In order to verify outputs of our model calculations and unusual mobile positioning
data, the distribution of homes calculated by the model was compared to the data
from the Estonian Population Register. It was impossible to retrieve respective data
of adequate quality and accuracy concerning the locations of work places.
There is a linear correlation (r ¼ 0.99) between the number of modeled homes
and the number of residents in the Population Register in Estonia’s 227 municipalities.
Local municipalities with larger populations also have more anchor points
in our model. (See Figure 9.) Due to regional and social peculiarities, there are a
number of deviations from the curve. Excluding the greater deviations in major
cities Tallinn, Tartu, Narva, Pa¨rnu, Kohtla-Ja¨rve, the correlation was r , 0.86.
Figure 8. The density of work-time anchor points in municipalities modeled using passive mobile
positioning data provided by EMT
20 Journal of Urban Technology
Downloadedby[MasarykovaUniverzitavBrne]at05:5216November2017
Table 2. Share of the Estonian home anchor points and population register data by countries
Place of residence Work-time anchors
Number Percentage Number Percentage
County Mobile positioning Population register 2007 Mobile positioning Population register 2007 Mobile positioning Mobile positioning
Harju county 181,969 536,152 38.4 39.4 187,308 39.5
Tartu county 59,997 146,749 12.7 10.8 61,017 12.9
Ida-Viru county 38,757 172,339 8.2 12.7 38,695 8.2
Pa¨rnu county 30,564 90,994 6.5 6.7 30,583 6.4
La¨a¨ne-Viru county 24,107 68,090 5.1 5.0 23,734 5.0
Viljandi county 21,391 55,547 4.5 4.1 20,977 4.4
Rapla county 15,934 37,341 3.4 2.7 13,966 2.9
Po˜lva county 15,677 32,062 3.3 2.4 14,999 3.2
Vo˜ru county 15,487 39,058 3.3 2.9 15,362 3.2
Jo˜geva county 14,511 36,698 3.1 2.7 13,963 2.9
Saare county 14,492 36,587 3.1 2.7 14,377 3.0
Valga county 14,269 35,007 3.0 2.6 13,872 2.9
Ja¨rva county 13,114 36,283 2.8 2.7 12,695 2.7
La¨a¨ne county 9809 28,249 2.1 2.1 9497 2.0
Hiiu county 3609 10,623 0.8 0.8 3549 0.7
UsingMobilePositioningDatatoModelLocationsMeaningfultoUsers21
Downloadedby[MasarykovaUniverzitavBrne]at05:5216November2017
In order to determine the deviations between two databases, the share of
modeled homes and registered persons of every municipality in Estonia (total
100 percent) were calculated more precisely. These two percentage-based datasets
were compared. There are more home anchor points based on the data of mobile
positioning than those registered in the Population Register in the near vicinity of
larger cities, especially Tallinn, e.g., municipalities such as Harku, Rae, and Loksa.
This pattern also applies to some cities such as Tartu and Vo˜ru and their nearby
surroundings. However, those deviations of homes were relatively small, with a
maximum of 0.64 percentage points in Harku. The most probable reason for
this is the intensive suburbanization that is taking place in Estonia (Tammaru
et al., 2009).
There are fewer home anchor points derived from mobile positioning data
than those registered in the Population Register in most of cities of Estonia, for
instance Tallinn, Narva, Sillama¨e, Kohtla-Ja¨rve, Maardu, Pa¨rnu, Elva, Valga,
Jo˜geva and in local governmental units farther from cities in central, western,
and southern Estonia. The differences in the share of the total number of
modeled homes from the Population Register reach 3.4 percentage points in
Tallinn and 2.9 in Narva. In Kohtla-Ja¨rve, Sillama¨e, and Maardu, the differences
are about one percentage point.
If generalized to county level, the greatest differences between mobile positioning
and Population Register data are found in North-East Estonia (Ida-Viru
County), as according to mobile-positioning-based model output, there are 4.5
percent fewer people living and working there than shown by the Population
Register. Poverty and related social problems may be one of the reasons why
mobile data is underappreciated in the North-Eastern Estonian industrial cities,
which results in the fact that many might not own a mobile phone or use
cheaper calling cards. Data from the current study originates from EMT, which
has both one of the best radio coverage and also above average price levels. In
spring 2008 (February-April) we ordered a special survey of mobile telephone
penetration and geographical share between operators from polling firm TNS
Figure 9. The correlation between the number of modeled home anchor points and the number of
registered persons in local municipalities in Estonia
22 Journal of Urban Technology
Downloadedby[MasarykovaUniverzitavBrne]at05:5216November2017
EMOR with 2000 respondents. The results showed that 95 percent of Estonians
have a mobile phone, and that figure was 96 percent in cities. EMT had the greatest
(70 percent) share of the market in South-Eastern Estonian rural Po˜lva county
(74 percent), and 59 percent in Vo˜ru county. The smallest share of EMT was in
Tallinn, with 39 percent, and northeastern Estonian counties Ida-Viru 37 percent
and La¨a¨ne-Viru 39 percent.
The other area with greater differences is Tartu County, as mobile positioning
data for that region shows a population that is 1.9 percentage points larger than
shown by the Population Register.
Discussion
Model and Sampling
The relatively complex model of home and work-time anchor points that was
developed during our practical traffic, commuting, and internal tourism research
has been proven to be somewhat scientifically acceptable. Nevertheless, the whole
system of inquiry and algorithm development need more exact analysis from
the point of view of geographical accuracy, social characteristics, and the interpretation
of an anchor point. All of the eight modeling stages need to be tested
with databases of different levels. Also, depending on the use of telephones, the
limits of standard deviations of our case by case set, temporal filtering times
and calls may vary. The important thing about our model is its applicability in
processing a large amount of data; our database had over 800 million data
entries, and performing so many inquiries required a completely different approach.
Our database and processing is based on PostgreSQL database manager inquiries,
which can manage even larger amounts of data. Our discussions during seminar
SPM2008 in Tartu showed that one of the greatest problems in using mobile positioning
and GPS data in geographical studies is the great amount of data involved,
which are not manageable with traditional programs and skills.
We believe that the model is accurate in finding routine places of personal
activity spaces. The comparison with the Population Register was relevant. There
is, however, still the possibility for systematic errors to occur, and these need to
be controlled in future. For example, perhaps commuters’ “most called” places
are on their everyday route from home to work and vice versa, not home or
work. Therefore the term of “meaningful place” is introduced from the beginning
and used parallel with anchor point and regularly visited place. We think that
because of specific passive positioning data, the term “meaningful place” fits
better than anchor point which normally has fixed and known location. We also
need to learn more about the meaning of quantitatively measured anchor points
because such a quantitative model is good for monitoring larger areas or sampling,
but is not good for analyzing personal activities, or places and their meaning.
Another important topic of discussion is the model’s compatibility with
different data sources. The model should be compatible with different sources,
since mobile operators’ infrastructures and information systems are different,
and data outputs may exist on different levels in the mobile network. In this
way, passive mobile positioning data may be obtained from terminals, positioning
servers, billing memory, or indirect sensors of radio location. Different network
standards differ in terms of data parameters for passive data outlets. For
example, the size of network cells and protocol for hand-over differs in Nokia
Using Mobile Positioning Data to Model Locations Meaningful to Users 23
Downloadedby[MasarykovaUniverzitavBrne]at05:5216November2017
and Ericsson network systems (antennae, information system, positioning server,
etc.) for GSMs used in Estonia’s largest networks. A lot of work must be done to
harmonize data outputs for CDMA, GSM, 3G, A-GPS and other types. The model
must also be tested with data from GPS-based tracking experiments, which are
very popular in studying space-time behavior.
A very important aspect to discuss is sampling. The data source is new, and
there is a need for future analysis concerning technical standards, sampling, and
methods (Ahas et al., 2008a). The sample is influenced by technical aspects such as
battery lifetime and the quality of radio coverage. For sampling, it is important to
be aware of regional and social differences in the penetration of phones and operators,
as well as peculiarities of phone use. For example, different ethnic groups
may use phones differently in Estonia because of the different costs of calls. The
activity of phone use is also determined by the age, profession, or travel behavior
of the respondent. All of these aspects do influence the location of call activities
and our geographical studies that depend on this data.
Geographical Accuracy
From the standpoint of the anchor point model, future discussions concerning
geographical accuracy are necessary. Today’s network cell accuracy makes it possible
to collect comparative results, but there are still problems in sparsely populated
rural areas where network cells and errors are greater. Confusion is also
created when work-time and home anchor points are placed in the same cell in
these big network cells; this was the case in 24 percent of the everyday anchor
points in this survey. This makes certain research tasks significantly more difficult
to fulfill. In terms of geographical distribution, the advantage of mobile networks
is that they follow persons, their living spaces and roads, and because of that are
geographically reasonably positioned (Ahas et al., 2008a).
In general, it is possible to raise the geographical accuracy of the passive positioning
data with the cooperation of operators, e.g., dividing cells into sectors or
using better positioning techniques for data collection. However, more accurate
data is always more expensive, and processing is more demanding. A technique
to manage the “tossing between antennas” is also necessary. Tossing means that
even if you are actually in one place, your phone may connect to different antennas
and swap (toss) them automatically, depending on the antennas’ workloads.
In GSM networks this poses a great problem for the Cell ID data and our
model. In the database, the location of the call is determined by the cell where it
began. In a 3G network, the phone is simultaneously connected to three antennas
and the calculations of location take all three into account. These technical problems
need to be dealt with very thoroughly, and models must be able to unify
data from different technical platforms.
Summary
The objective of the current study was to develop and to test the model for determining
meaningful locations of mobile phone users as locations of homes and
work-places. The database of Estonia’s largest mobile operator, EMT, which
covers the whole country, was used for modeling. In our opinion the results of
the model calculations were quite good; this is supported by comparisons of
24 Journal of Urban Technology
Downloadedby[MasarykovaUniverzitavBrne]at05:5216November2017
our modeling results to those of the Population Register. Differences between the
data of mobile positioning and population register were greater in socially
unstable areas and regions affected by intensive urban sprawl. In a broader
sense, the mobile positioning based data and methods may be the best or only
available source for monitoring population processes in such unstable or
rapidly developing regions.
We conclude that it is possible to monitor a population’s geography and
mobility by using the passive databases of mobile operators. However, the development
of a perfect system demands a great deal of additional work, as the input
data and model needs to be standardized for different sources and conditions.
In terms of geographical research, the current methodology is promising, and
we sought to distinguish new features concerning different location points of the
majority of population with relatively low research costs. This makes the database
attractive for different research perspectives and applications. There is great
potential for the development of real-time monitoring tools and geographical
applications. There is growing interest in such geography of regularly visited
places in geography for tourism development, traffic management, and urban
planning applications. The information technology commercial sector can also
use the model to develop and personalize mobile services; to develop personalized
and location-aware advertisements; and to optimize radio coverage and
other network services.
Acknowledgments
The authors wish to thank EMT Ltd., Ericsson Ltd., Positium LBS and all the
people who participated in the experiment for their cooperation. The project
was funded by Target Funding Project No. 0182143s02 of the Ministry of
Education and Science and Grant of Estonian Science Foundation No. ETF7562
and Estonian Information Technology Foundation (EITSA).
Using Mobile Positioning Data to Model Locations Meaningful to Users 25
Downloadedby[MasarykovaUniverzitavBrne]at05:5216November2017
Bibliography
R. Ahas, A. Aasa, A. Roose, U¨ . Mark, and S. Silm, “Evaluating Passive Mobile Positioning Data for
Tourism Surveys: An Estonian Case Study,” Tourism Management 29 (2008a) 469–486.
R. Ahas, A. Aasa, S. Silm, R. Aunap, H. Kalle, and U¨ . Mark, “Mobile Positioning in Space-Time Behavior
Studies: Social Positioning Method Experiments in Estonia,” Cartography and Geographic Information
Science 34:4 (2007a) 259–273.
R. Ahas, J. Jauhiainen, S. Silm, M. Nuga, A. Ta¨hepo˜ld, E. Saluveer, and K. Kivi, Cooperation Perspectives
between Municipalities of the Urban Region of Tallinn (Tallinnaja u¨mbritsevate omavalitsuste koosto¨o¨vo˜imalused
ja perspektiivid valglinnastumise kontekstis) (Tartu: University of Tartu, Institute of Geography, 2007b).
R. Ahas and U¨ . Mark, “Location Based Services: New Challenges for Planning and Public Administration?’’
Futures 37 (2005) 547–561.
R. Ahas, E. Saluveer, M. Tiru, and S. Silm, “Mobile Positioning Based Tourism Monitoring System: Positium
Barometer,” in J. O` Connor, W. Ho¨pken, and U. Gretzel, eds., Information and Communication
Technologies in Tourism (Wien, New York: Springer, 2008b), pp. 475–485.
Y. Asakura and E. Hato, “Tracking Survey for Individual Travel Behavior Using Mobile Communication
Instruments,” Transportation Research Part C 12 (2004) 273–291.
K.W. Axhausen, A. Zimmermann, S. Scho¨nfelder, G. Rindsfu¨ser, and T. Haupt, “Observing the
Rhythms of Daily Life: A Six-Week Travel Diary,” Transportation 29 (2002) 95–124.
X.L. Chen, H.M. Zhao, and P.X. Li, “Remote Sensing Image-Based Analysis of the Relationship between
Urban Heat Island and Land Use/Cover Changes,” Remote Sensing of Environment 104:2 (2006) 133–146.
M. Dijst, “Two-Earner Families and Their Action Spaces: A Case Study of Two Dutch Communities,”
GeoJournal 4:3 (1999) 195–206.
K.V. Du¨rckheim, “Untersuchungen zum Gelebten Raum: Erlebniswirklichkeit und ihr Versta¨ndnis
Systematische Untersuchungen II,” Neue Psychologische Studien 6 (1932) 383–480.
N. Eagle and A. Pentland, “Reality Mining: Sensing Complex Social Systems,” Personal and Ubiquitous
Computing 10:4 (2005) 255–268.
Eurobarometer, E-Communications Household Survey, Fieldwork November – December 2006, Special
Eurobarometer 274/Wave 66.3 (TNS Opinion & Social, 2007), ,http://ec.europa.eu/public_opinion/
archives/ebs/ebs_293_full_en.pdf.
Ericsson, Home (2008) ,www.erricsson.com. Accessed April 2008.
European Parliament, Council of 12 July 2002, Directive 2002/58/EC (Directive on Privacy and
Electronic Communications) ,http://eur-lex.europa.eu/LexUriServ/site/en/oj/2002/1_201/1_201
2002073en00370047.pdf. Accessed April 2008
M. Flamm, Comprendre le Choix Modal – Les De´terminants des Pratiques Modales et des Repre´sentations
Individuelles des Moyens de Transport doctoral dissertation (Lausanne: Ecole Polytechnique Fe´de´rale
de Lausanne, 2004).
M. Flamm and V. Kaufmann, “The Concept of Personal Network of Usual Places as a Tool for
Analyzing Human Activity Spaces: A Quantitative Exploration,” paper presented at 6th Swiss
Transport Research Conference, Ascona 2006).
G. Gartner, “Location-Based Mobile Pedestrian Navigation Services: The Role of Multimedia
Cartography,” paper presented at UPIMap2004 (Tokyo, Japan, 2004).
R.G. Golledge, “Cognition of Physical and Built Environments,” in T. Garling and G. Evans, eds., Environment,
Cognition and Action: A Multidisciplinary Integrative Approach (New York: Oxford University Press,
1990).
R.G. Golledge and R.J. Stimson, Spatial Behavior: A Geographical Perspective (New York: The Guilford
Press, 1997).
R.Herman,“MonitoringTrafficinRealTimeFromtheComfortofYour Car,”Computerworld40:39(2006)31.
T. Ha¨gerstrand, “What about People in Regional Science?’’ Papers and Proceedings of the Regional Science
Association 24 (1970) 7–21.
O. Ja¨rv, E. Saluveer, and R. Ahas, Analysis of Traffic Distribution and Frequencies in Kose-Vo˜o˜bu ja Vo˜o˜buMa¨o
Units of E263 Tallinn-Tartu-Luhamaa Highway (E263 Tallinna-Tartu-Luhamaa maantee Kose-Vo˜o˜bu ja
Vo˜o˜bu-Ma¨o lo˜ikude Liiklusuuring Mobiilpositsioneerimise Abil) (Tartu: Positium LBS, 2007).
J.B. Kang, W. Welbourne, B. Stewart, and G. Borriello, “Extracting Places from Traces of Locations,”
in Proceedings of the 2nd ACM International Workshop on Wireless Mobile Applications and Services on
WLAN Hotspots (Philadelphia, PA, USA: ACM Press, 2004).
T. Kopomaa, “Speaking Mobile: Intensified Everyday Life, Condensed City,” in S. Graham, ed., The
Cybercities Reader (London: Routledge, 2004).
26 Journal of Urban Technology
Downloadedby[MasarykovaUniverzitavBrne]at05:5216November2017
E. Kutter, “Aktionsbereiche des Stadtbewohners: Untersuchungen zur Bedeutung der territorialen
Komponente im Tagesablauf der Sta¨dtischen Bevo¨lkerung,” Archiv fu¨r Kommunal- wissenschafte 13
(1973) 69–85.
M.P. Kwan, “Interactive Geovisualization of Activity-Travel Patterns Using Three-Dimensional
Geographical Information Systems: A Methodological Exploration with a Large Data Set,” Transportation
Research Part C Emerging Technologies 8:1–6 (2000) 185–203.
R.B. Mitchell and C. Rapkin, Urban Traffic: A Function of Land Use (New York: Columbia University
Press, 1954).
D.M. Mountain, Exploring Mobile Trajectories: An Investigation of Individual Spatial Behavior and Geographic
Filters for Information Retrieval doctoral dissertation (London: Department of Information Science,
City University London, 2005) PhD Thesis.
D.M. Mountain and J. Raper, “Modeling Human Spatio-Temporal Behavior: A Challenge for LocationBased
Services,” in Proc. of the 6th International Conference on GeoComputation, (Brisbane, Australia:
University of Queensland, 2001).
P. Nurmi and S. Bhattacharya, “Identifying Meaningful Places: The Nonparametric Way,” J. Indulska et
al., eds, in Proceedings of the 6th International Conference on Pervasive Computing (Pervasive’08), (Sydney:
Springer-Verlag, 2008), 111–127.
P. Nurmi and J. Koolwaaij, “Identifying Meaningful Locations,” paper presented at 3rd Annual
International Conference on Mobile and Ubiquitous Systems: Networks and Services (MobiQuitous),
San Jose, CA, 2006).
R. Oldenburg, The Great Good Place: Cafes, Coffee Shops, Community Centers, Beauty Parlors, General Stores,
Bars, Hangouts, and How They Get You Through the Day (New York: Paragon House, 1989).
Positium LBS (2008) ,www.positium.ee. Accessed April 2008
C. Ratti, “Mobile Landscape: Graz in Real Time,” in Proceedings of 3rd Symposium on LBS & TeleCartography,
(Vienna Vienna University of Technology, 2005).
J.G. Raymer, G. Abel, and P.W.F. Smith, “Combining Census and Registration Data to Estimate Detailed
Elderly Migration Flows in England and Wales,” Journal of the Royal Statistical Society Series AStatistics
In Society 170:4 (2007) 891–908.
J. Reades, F. Calabrese, A. Sevtsuk, and C. Ratti, “Cellular Census: Explorations in Urban Data
Collection,” IEEE Pervasive Computing 6:3 (2007) 30–38.
T. Roos, P. Myllyma¨ki, and H. Tirri, “A Statistical Modeling Approach to Location Estimation,” IEEE Transactions
on Mobile Computing 1:1 (2002) 59–69.
R.C. Schank and R.P. Abelson, Scripts, Plans, Goals, and Understanding: An Inquiry into Human Knowledge
Structures (Hillsdale: Lawrence Erlbaum, 1977).
B. Schwarze and S. Scho¨nfelder, ArcView-Extension VISAR: Visualisierung von Aktionsra¨umen, Version 1.6
(Zu¨rich: Institut fu¨r Verkehrsplanung, Transporttechnik, Stassen- und Eisenbahnbau [IVT], 2001).
S. Scho¨nfelder and K.W. Axhausen, Structure and Innovation of Human Activity Spaces (Arbeitsbericht:
Verkehrs- und Raumplanung 258, 2004).
M. Sheller and J. Urry, “The New Mobilities Paradigm,” Environment and Planning A 38:2 (2006)
207–226.
N. Shoval, “Sensing Human Society,” Environment and Planning B 34 (2007) 191–195.
S. Silm, R. Ahas, and M. Nuga, “Gender Differences in Space-Time Mobility of Dwellers from New
Residential Areas in the Suburbs of Tallinn,” (Tartu: Study Report, Department of Geography, University
of Tartu, 2008).
S. Silm and O. Ja¨rv, Summerhouses, Users and Usage Frequencies in Saku County (Saku valla suvilate kasutuse
uuring) (Tartu: Positium LBS, 2007).
S. Silm, O. Ja¨rv, and R. Ahas, Urban Sprawl in Kuressaare Urban Region (Valglinnastumine ja sellest la¨htuvad
mo˜jud Kuressaare linnaregioonis: esmane analu¨u¨s ja ettepanekud planeerijatele) (Tartu: Positium LBS, 2007).
J.E. Spinney, “Mobile Positioning and LBS Applications,” Geography 88 (2003) 256–265.
T. Tammaru, K. Leetmaa, S. Silm, and R. Ahas, “Temporal and Spatial Dynamics of the New Residential
Areas around Tallinn,” European Planning Studies 17:3 (2009) 423–439.
V. Vidakovic, “De Toedeling van Reistijd aan Activiteiten: Eerste Resultaten van Waarnemingen in
Amsterdam,” Verkeerskunde 32 (1981) 442–444.
J.L. Wolf, R. Guensler, and W.H. Bachmann, “Elimination of the Travel Diary: An Experiment to Derive
Trip Purpose from GPS Travel Data,” Journal of the Transportation Research Record 1768 (2001) 125–134.
Y. Zhao, “Standardization of Mobile Phone Positioning for 3G Systems,” IEEE - Communications
Magazine 40:7 (2002) 108–116.
C. Zhou, D. Frankowski, P. Ludford, S. Shekhar, and L. Terveen, “Discovering Personally Meaningful
Places: An Interactive Clustering Approach,” ACM Transactions on Information Systems (TOIS) 25:3
(2007) 1–31.
Using Mobile Positioning Data to Model Locations Meaningful to Users 27
Downloadedby[MasarykovaUniverzitavBrne]at05:5216November2017