Professional Documents
Culture Documents
O13 Brandtzaeg Et Al (2011) - Kmeans
O13 Brandtzaeg Et Al (2011) - Kmeans
Abstract
The purpose of this study is to better understand the digital divide by identifying the variety of ways in which people in Europe use the
Internet. First, by using cluster analysis on survey data (N= 12,666/age: 16–74 years) from Eurostat on Internet usage in Norway, Sweden,
Austria, the UK, and Spain, we identified five user types: Non-Users (42%), Sporadic Users (18%), Instrumental Users (18%),
Entertainment Users (10%), and Advanced Users (12%). These user types differ in their distributions over country, age, access, household
members, and gender. An alarming finding is that 60% of the population was found to be either Non-Users or Sporadic Users, which
reflects a large digital divide in Europe. Second, we conducted a logistic regression to identify the predictors for different user types.
We found on a cross-national level that age and Internet access are the most salient predictors, whereas gender and household seems to be
less relevant. However, the amount of variance explained differs between countries. We also suggested a future increase in the digital divide
between the identified user types—a user type divide. The user typology and the identified predictors might help researchers, practitioners,
and decision makers to better understand Internet users and the multi-complex variations among individuals and countries. This knowledge
will also serve as a means to understand the digital divide by providing a more nuanced perspective on Europeans’ unequal usage of the
Internet and participation in an increasingly digital society.
& 2010 Elsevier Ltd. All rights reserved.
1071-5819/$ - see front matter & 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ijhcs.2010.11.004
124 P.B. Brandtzæg et al. / Int. J. Human-Computer Studies 69 (2011) 123–138
One of the most comprehensive studies with regard to 2.2. Limitations of previous research and contributions
Internet user types conducted in the European context is by of this study
Ortega Egea et al. (2007), which used a representative
questionnaire dataset from European countries (EU15) for There are several existing research gaps that this study will
the purpose of international comparison. Data were gath- try to overcome. First, a comprehensive understanding of
ered in November 2002, and a cluster analysis identified the Internet use is a prerequisite for overcoming the digital divide.
following five types of European Internet users: So far, few studies fully understand the multi-complex
variations among individuals that constitute the digital divide
(e.g., Ortega Egea et al., 2007; Brandtzæg, 2010). This study
(1) Laggards (16%)—Occasional and infrequent use of
aims to show that users can be usefully separated into
Internet services and no use of eGovernment services.
distinguishable segments based on their actual Internet
Laggards rarely use the Internet for private purposes.
Most Laggards were found in Germany, France, and behavior. Classifying Internet usage in terms of distinct user
types will therefore help us to more fully understand the
Ireland.
differences in adoption rate and online participation among
(2) Confused and adverse (2%)—High variability, but gen-
both individuals and countries. Second, only five studies have
erally low usage of Internet. This user type shows
previously identified user types across countries, few are
confusion about Internet services. This category rarely
academic (Brandtzæg, 2010), and most of them use outdated
uses the Internet for private purposes or for contacting
empirical data compared to the present study. The usage
authorities, and they are mainly found in the United
patterns of Internet services in Europe have probably changed
Kingdom and Austria.
(3) Advanced Users (16%)—Frequent use of eGovernment since 2002, when the empirical data published by Ortega Egea
et al. (2007) were collected. To overcome the digital divide in
services, not only for administrative tasks (e.g., to search
Europe, a deeper, more contemporary understanding of
for administrative information, to fill out forms, or to
Internet usage among European citizens is needed. Third,
carry out administrative transactions) but also for other
few studies have examined the predictors of different user types
purposes. Advanced Users are the most frequent online
(Brandtzæg, 2010).
shoppers. Countries with most Advanced Users were the
Finally, this study aims to go beyond other similar studies
UK, Holland, and the Nordic countries.
(e.g., Johnsson-Smaragdi, 2001; Selwyn et al., 2005; Horrigan,
(4) Followers (19%)—Use the Internet quite frequently but
not on a daily basis. Followers use eGov-services, 2007; Ortega Egea et al., 2007) by covering a broader spectrum
of usage and content preferences, including entertainment
although not as frequently as Advanced Users and do
activities. Entertainment usage is taken into account among
not shop online. Most Followers were found in Holland
some previous studies of user types (e.g., Heim et al., 2007) but
and Denmark.
not in those that focus on the adult population. Entertainment
(5) Non-Internet Users (44%)—The largest group in the
usage online is thought to have increasing importance in the
sample is characterized by non-usage of the Internet.
information society for both child and adult populations
Countries with a majority of Non-Internet Users were in
(Brandtzæg et al., 2003; Taylor, 2006).
the southern part of Europe, such as Spain, Greece,
Portugal, and Italy.
2.3. Research questions
This user typology reflects not only how different user By using a unique set of data covering European Internet
groups use the Internet in various ways, but also how usage, the goal of this study is to identify and explain a
dissimilar is the potential of user types to exploit the benefits typology of Internet users that offers a better understanding
of the Internet. The user typology identified by Ortega Egea of the various types of usage that constitute the digital
et al. (2007) is partly in line with the unified Media User divide. As a result, we aim to answer the following research
Typology (MUT) model presented by Brandtzæg (2010), questions:
which identified eight new media users: (1) Non-Users, (2)
Sporadics, (3) Debaters, (4) Entertainment Users, (5)
Lurkers, (7), Socializers, and (8) Advanced Users, where
RQ1: What are the different user types in Europe in
regard to Internet usage?
all, except Debaters (people using Internet for discussion)
and Socializers (people using Internet for socialize with
RQ2: How can we explain the different user types within
countries and across different countries?
friends and family), are identified based on their general
J What is the relationship within and between countries,
Internet behavior. Debaters and Socializers are more
genders, age groups, types of household, and levels of
common to be found among users in blogs and social
access to technology with respect to different
networking sites in particular, which is an increasingly
user types?
common user behavior. However, blogs and social network-
ing sites were not mainstream in 2005–2006, when the
data from this study were collected, and certainly not The motivations for the use of (1) countries, (2) access (3)
in 2002. gender (4) age, and (5) members in the household as
126 P.B. Brandtzæg et al. / Int. J. Human-Computer Studies 69 (2011) 123–138
Table 1
Method overview for each country.
Total 44,102
a
Office for National Statistics, Social Survey Division, ONS Omnibus Survey, Internet Access Module, April, July and October, 2004 and February 2005 [File:
2004-05_m330_dataset]. Colchester, Essex: UK Data Archive, Distributed by UK Data Archive, University of Essex, Colchester. November 2006. SN: 5542.
predictors for the various Internet usage or user types are as moderate use of new media (e.g., Germany, Switzerland, and
follows. Austria), heavy use of both TV and new media (only the UK
belongs to this group), and pioneers of new media (the Nordic
(1) Country: Ortega Egea et al. (2007) found quite large countries and the Netherlands). The selection of participat-
differences in user types between countries (see back- ing countries in our study was motivated by a desire to
ground section). include countries characterized by low, medium, and high
(2) Access: The form of Internet access (no access, broad- Internet use as well as the availability of micro-data from
band, or narrowband) is an obvious predictor for national statistical agencies. Thus, we selected countries
different types of usage (see Table 2 for overview of with relatively low (Spain), medium (Austria and UK), and
various accesses among participants in our sample). high (Norway and Sweden) use of the Internet. Table 2
(3) Gender: Numerous studies have found a large gender below shows the penetration rates in terms of various forms
divide in the use of Internet (e.g., Hargittai, 2010; of Internet access in the different countries.
Dholakia, 2006; Bimber, 2000) and in technology adoption We received micro-data2 on Internet use by household and by
contexts in general (e.g., Venkatesh et al., 2003). individuals from the statistical offices of five European
(4) Age: A generational divide is identified between older countries. Table 1 gives an overview of the offices that provided
and younger Internet users, where the older users are the data, the period of the data collection, the number of
often found to lag behind, both in usage and access (see respondents, and the interview method used for each country.
overview in Karahasanović et al. (2009). The original total sample size was N=44,102. Tables A1
(5) Household: With regard to household composition, it is a and A2 in Appendix A give an overview of the different
common opinion that having children in the household countries’ sample characteristics compared to real country
increases the probability that the household will acquire population characteristics. To adjust for the non-representa-
computers and Internet access (Heim et al., 2007). tive distribution of countries (different sample size) and some
gender skews in the Eurostat data (compared to the population
3. Method data), a random selection of participants was performed so
that a country’s population and gender distribution were
We used the data from the European community survey on mirrored in the new sub-sample (N=12,666). The new,
ICT usage in households and by individuals from 2004 to 2006. adapted sample is shown in Table 2. The new and final sample
The collection of the raw data was coordinated by the EU’s (N=12,666) was compared and verified with Eurostat data on
statistical office—Eurostat.1 The survey was first introduced in age and gender distribution in the respective countries (see
2002. It currently covers most of the EU countries and some of Tables A1 and A2 in Appendix A).
the EFTA countries and aims to collect harmonized and
comparable information on European citizens’ access to and
use of the Internet. 3.2. Scope
This study uses the data on individuals aged 16–74 years
3.1. Countries analyzed and households with at least one member in the same age
group. The statistical unit is the individual respondent in a
Livingstone et al. (2001) classified European countries household. The survey maps overall usage of the Internet for
based on their use of both old (TV, radio, and newspapers) individuals and households, and is divided along variables
and new media (PC and Internet) into the following four such as family type, age, gender and form of Internet access.
groups: low use of new media (e.g., Spain, Italy, and France),
2
Micro-data are data on the characteristics of units of a population, such
1
http://epp.eurostat.ec.europa.eu. as individuals or households.
P.B. Brandtzæg et al. / Int. J. Human-Computer Studies 69 (2011) 123–138 127
Table 2
Sample characteristics (in %) in the selected countries.
Note: The final new sample N=12,666. See the proportional sampling in Tables A1 and A2 in Appendix A. N/A =not applicable.
As shown in Table 1, the respondents’ Internet usage is 3.4. Identifying Internet user types
measured using a self-assessment approach, i.e., the respon-
dent simply indicates whether he/she use various Internet To answer RQ1, to identify the user typology, we
activities. The complete Eurostat methodology and ques- conducted a K-means cluster analysis of people’s usage of
tionnaire can be found online.3 technology. A cluster analysis is a robust technique for
classifying similar objects into different groups. More
3.3. Harmonization of data precisely, a cluster is a group of relatively homogeneous
cases or observations. Objects in a cluster are similar to each
The questions concerning Internet usage and access asked in other, and dissimilar to objects in other clusters. Cluster
different countries were mainly subsets of the ‘‘Eurostat model analysis is the most commonly used technique in studies that
for a community survey on ICT usage in households and by identify media user types (Brandtzæg, 2010). K-means
individuals, 2005.’’ Although the questionnaire items differed cluster analysis (see for example Tan et al., 2006) is
among countries, there was a group of core items used by all. recommended when the number of entities (persons) in
The data were received in different formats such as Excel the analysis is high (more than 1000). In a K-means cluster
(Sweden), plain text (Spain), and SPSS (Norway, UK and analysis, K denotes the number of clusters and is an input
Austria), and they were coded differently. First, all files were parameter. Moreover, all variables should be within the
formatted into an SPSS format. Next, all variable labels were same range when entered into the analysis; in our dataset, all
translated into English by a professional translator, and all variable values were 0 or 1 (see Table 3).
variables that were based on the same question were re-coded We conducted a K-means score analysis of the data (see
into a common format when possible, which included recoding Table 3). First, we identified a set of variables that was used
from multiple choice responses into ‘‘Yes/No’’ responses, in the analysis. A subset of 29 items appeared in the datasets
giving ‘‘Yes/No’’ responses the same numerical codes (No=0, from all five countries. These items covered questions on
Yes=1), and coding Age group, Gender, Number of members Internet usage, regardless of what type of Internet access the
in household and Access to PC and Internet into variables with respondent had. Six of these variables had a very low
common categories. Finally, all variables that were assessed as frequency of ‘‘Yes’’ responses, so they were thought to be
‘‘the same’’ were given identical names and merged into one unsuitable for analysis and were deleted from the common
SPSS file for further analysis. set, which was then reduced to 23 clustering variables.
Second, to determine the number of clusters, we followed
3
See http://epp.eurostat.ec.europa.eu/cache/ITY_SDDS/FR/isoc_reg_ the procedures suggested by Ray and Turi (1999). As a
base.htm. result, we identified five clusters denoting five Internet user
128 P.B. Brandtzæg et al. / Int. J. Human-Computer Studies 69 (2011) 123–138
Table 3
Mean scores within each cluster.
Variable Cluster
1 2 3 4 5
Have you used the Internet daily during the past three months? 0 0.11 0.74 0.51 0.83
Have you used the Internet at home during the past three months? 0 0.44 0.87 0.87 0.91
Have you used the Internet at your workplace outside your home during the past three months? 0 0.27 0.35 0.51 0.76
Have you used the Internet in an educational institution during the past three months? 0 0.08 0.30 0.06 0.16
Have you used the Internet in someone else’s home during the past three months? 0 0.19 0.28 0.11 0.39
Have you used e-mail during the past three months? 0 0.42 0.94 0.95 0.99
Have you taken part in a chat session during the past three months? 0 0.08 0.57 0.07 0.38
Have you searched for information about goods or services during the past three months? 0 0.45 0.88 0.93 0.98
Have you utilized services related to travel or hotels during the past three months? 0 0.13 0.33 0.75 0.96
Have you listened to or watched Internet radio or TV during the past three months? 0 0.11 0.71 0.08 0.46
Have you played or downloaded games or music online during the past three months? 0 0.12 0.82 0.07 0.44
Have you used net-banking services during the past three months? 0 0.08 0.42 0.50 0.76
Have you purchased or ordered goods or services during the past three months? 0 0.05 0.28 0.42 0.92
Have you taken any continuing education courses during the past three months? 0 0.04 0.13 0.12 0.34
Have you taken other educational courses in connection with work possibilities during the past 0 0.05 0.10 0.17 0.47
three months?
Have you purchased or ordered household products during the past 12 months? 0 0.01 0.04 0.08 0.29
Have you purchased or ordered films or music during the past 12 months? 0 0.02 0.13 0.09 0.50
Have you purchased or ordered books, newspapers, magazines or teaching materials during the 0 0.02 0.1 0.12 0.50
past 12 months?
Have you purchased or ordered clothes or sports articles during the past 12 months? 0 0.03 0.1 0.14 0.35
Have you purchased or ordered electronic equipment during the past 12 months? 0 0.01 0.07 0.07 0.30
Have you purchased or ordered travel or hotel rooms during the past 12 months? 0 0.02 0.05 0.22 0.81
Percentage of cases/clusters (%) 42 18 10 18 12
Note: N=12,666. Different font styles are used to enhance the readability of the table: italics is used for cluster means equal or less than 0.25 and bold is used
for cluster means equal or greater than 0.75.
types, with the details given in Fig. A1 in Appendix A. National samples: Total sample divided into five sub-
Lastly, we calculated mean scores for each variable for the samples according to participant’s country (Norway,
five clusters and described them in terms of the user behavior Austria, Sweden, Spain and UK—see Table 2).
typical of each cluster. The data were analyzed with SPSS for
Windows version 15 (SPSS Inc., Chicago, IL, USA). For each sample defined above, five logistic regression
Table 3 shows the mean score within each cluster (because analyses were carried out, one for each user type, where the
all raw data values are either 0 or 1, the ‘‘mean score’’ is also dependent variable was cluster membership (scored 0 or 1).
the proportion of cluster members responding ‘‘Yes’’ to The first analysis aimed at differentiating the Non-Users
the item). from the rest of the population. The next four analyses
aimed to differentiate the other four types from each other,
leaving out the Non-Users from the analyses.
3.5. Explaining and predicting Internet usage The following variables were entered as independent vari-
ables into the logistic regression at the same time: (i) access to
To answer RQ2, we wanted to extend our existing PC and Internet at home, (ii) number of members in house-
knowledge of Internet usage among European citizens by hold, (iii) gender, and (iv) age. Access to PC and Internet was
investigating the forces that drive the adoption of the grouped in the following categories: (1) no access, (2) access to
Internet and different types of adoption and use. More PC with no Internet connection, (3) access to PC and
specifically, we investigated relationships between countries, narrowband Internet connection, and (4) access to PC and
genders, age groups, access levels, and household sizes broadband Internet connection. Members in household cate-
pertaining to various Internet user types. We first identified gories were as follows: (1) household with one member, (2)
and characterized the user types and then conducted a household with two members, (3) household with three
logistic regression to identify the predictive factors for members, and (4) household with four or more members.
different user types. The age categories were as follows: 16–24, 25–34, 35–44, 45–54,
The following samples were used in the analysis: 55–64, and 65–74 years (see Table 2 for an overview). As seen
in Tables 4 and 5, several of the independent variables were
Total sample: A sample (N =12,666) of all participants correlated. Because the dependent variables – user type
selected to adjust for population size and gender (see membership – were dichotomous, logistic regression analyses
Table 2 and Tables A1 and A2 in Appendix A). were conducted instead of linear regression analyses.
P.B. Brandtzæg et al. / Int. J. Human-Computer Studies 69 (2011) 123–138 129
4. Results 60
Sporadic users
Instrumental users
50
We present the analyses in the following order: the Non users
% within country
Advanced users
interpretation of the K-means cluster analysis identifying 40 Entertainment users
major user types (Section 4.1), user types distributed across
the selected European countries (Section 4.2), and a logistic 30
regression analyzing the different predictors of various types 20
of users (Section 4.3). The same analysis conducted for the
entire sample was also carried out for each country to 10
explore the differences between countries (Section 4.4).
0
Norway Sweden Austria UK Spain
4.1. User types Country
Based on the K-means cluster analysis, we identified the Fig. 1. User types and distribution per country.
following five types of Internet users, labeled and interpreted
in terms of the Media User Typology suggested by
Brandtzæg (2010), described in the background section Table 4
(Section 2): Spearman’s rho correlations among the independent variables in this study.
4.2. Distribution of user types among the participating 4.3. The predictors of different Internet user types
countries
We conducted a logistic regression analysis to investigate
Fig. 1 reveals that all user types are identified in all factors that may predict different types of users. Table 4
countries, but that there are substantial differences among shows that there were substantial correlations between the
130 P.B. Brandtzæg et al. / Int. J. Human-Computer Studies 69 (2011) 123–138
Table 5
Logistic regression analysis with the different user types as dependent variables across all countries (N=12,666).
Note: ‘‘Household’’ refers to the number of members in the household. ‘‘Access’’ refers to access to Internet and PC at home. In Table 6, we present the results
within each separate country for different user types.
independent variables, especially between access, age, and Non-User type (p=0.000). The model as a whole explained
members in a household. 58.1% (Nagelkerke R-squared) of the variance in Non-Users.
Table 5 presents the results of the logistic regression analysis, As shown in Table 5, being female increases the odds of
explaining the predictors for the different user types across being a Non-User by a factor of 1.35. Belonging to an older
different countries. The particular type of user (i.e., Non-User, age category increases the probability by a factor of
Sporadic User, Advanced User, Instrumental User, or Enter- 2.22. Increasing the access category by one unit decreases
tainment User) is the dependent variable. The independent the probability by a factor of 0.244. Belonging to a house-
variables are listed in the first column. For each user type, we hold category with more members increases the probability
report the results from the total sample and the Nagelkerke by a factor of 1.56. Comparing the Wald Chi-Square
R-square values, which provide an indication of the amount of statistics shows that Access and Age are the most significant
variation in the dependent variable (user type) explained by the predictors; this measure indicates the statistical significance of
model (from a minimum value of 0 to a maximum of the variable’s contribution to predicting the dependent
approximately 1). The ‘‘B’’ values in the left column in Table 5 variable.
are equivalent to the B values obtained in the multiple regression
analysis; B values, positive or negative, tell you about the 4.3.2. Predicting Sporadic Users
direction of the relationship. The ‘‘odds ratio’’ column to the As shown in Table 5, all independent variables contributed
right in Table 5 gives the factor by which the odds of a user significantly to the prediction of the Sporadic User type
belonging to a specific user type increase when the value of a (p=0.000). However, while the independent variables account
predictor are increased by one unit. This statistic reflects the for 58.1% of the variance among Non-Users, they only
effect size of the variable. The results are explained in more detail account for 25% (Nagelkerke R-squared) of the variance
below. The more descriptive analyses of user types and among Sporadic Users. Females tend to be Sporadic Users
predictors are to be found in Figs. A2–A5 in Appendix A. more often than men, and gender (being female) increase the
odds ratio by 1.37, being older increases the probability of
4.3.1. Predicting Non-Users being a Sporadic User by an odds ratio of 1.23, as does living in
All of the independent variables (access, age, gender, a larger household (odds ratio=1.3). For more about gender
household) contributed significantly to the prediction of the differences, see Fig. A2 in Appendix A.
P.B. Brandtzæg et al. / Int. J. Human-Computer Studies 69 (2011) 123–138 131
4.3.3. Predicting Entertainment Users highest penetration of broadband connections, while in the
As shown in Table 5, all of the independent variables UK and Austria narrowband is more common. Spain also
contributed significantly to the prediction of being an has the lowest level of access at home (see Table 2).
Entertainment User, except household. The model as a
whole explained 18% (Nagelkerke R-squared) of the 4.4.2. Sporadic Users
variance in Entertainment Users. Age and access are the As shown in Table 6, the major trends are similar to those
most significant predictors (see Wald in Table 5). for the analysis including all countries, but the predictor
variables account for more of the variance in the dependent
4.3.4. Predicting Instrumental Users variables for Sweden, Spain and UK than for Norway and
Age, access and gender contribute significantly to the Austria. In Spain, the model as a whole explained 29%
prediction of the Instrumental User type (p =0.000), but (Nagelkerke R-squared) of the variance in Sporadic Users.
household does not. Access and age are the most significant In Norway and Austria, access is the only significant
predictors (Wald). The model as a whole explained an predictor. The role of gender varies somewhat across
estimated 8% (Nagelkerke R-squared) of the variance in countries and is the most important in Spain, whereas the
Instrumental Users, which is little compared to the predic- effect of the number of household members is most
tions of variance in Non-Users. Slightly more females significant in the UK and Sweden.
than males tend to be Instrumental Users (see Fig. A2 in
Appendix A), and higher levels of access, being female and
4.4.3. Entertainment Users
older increase the odds ratio of being an Instrumental User
Table 6 shows that the role of gender is not significant in
between 1.2 and 1.4.
Norway, but it is highly significant in Sweden, UK and
Spain. The overall picture is that more males than females
4.3.5. Predicting Advanced Users
tend to be Entertainment Users (see Fig. A2 in Appendix A),
All independent variables contributed significantly to the
and younger users have a significant higher probability than
prediction of the Advanced User type (p =0.000). As shown
older users of being Entertainment Users in all countries
in Table 5, the model as a whole explained 8% (Nagelkerke
(p o 0.001). Moreover, a high level of access predicts being
R-squared) of the variance in Advanced Users. These results
an Entertainment User, but not significantly in Austria. To
are very much the opposite of the results for the Sporadic
some extent, the number of household members predicts
Users; being female, older and living in households with
being an Entertainment User in Norway, but not in the other
more members decreases the probability of being an
countries.
Advanced User, while greater Internet access increases this
probability of being an Advanced User (odds ratio =1.97).
4.4.4. Instrumental Users
4.4. Differences between countries Fig. A3 shows that being female and older adult increases
the probability of being an Instrumental User; however, the
The same regression analysis as conducted for the entire role of gender is not significant in Sweden and Spain. In
sample was carried out for each separate country, to explore Norway and Sweden the model as a whole explained,
the various predictors for different user types within the respectively, 17% and 11% (Nagelkerke R-squared) of
selected countries. The detailed results are given in Table 6. the variance in Instrumental Users. The model is not very
In general, Table 6 shows that the research model explains good at explaining the Instrumental Users among the other
the variance in the dependent variable to varying degrees countries. Further, while in the total analysis (Table 5) more
depending on the specific country. The model as a whole members of the household have no effect on the probability
explained Non-Users best in Norway with an estimated 60% of being an Instrumental User, this variable increases the
(Nagelkerke R-squared) of the variance, while Sporadic probability in Norway and the UK. Access increases the
Users was best explained in Spain (Nagelkerke R- probability in Norway, Austria and Spain. ‘‘Narrowband
squared = 28.6%), Entertainment Users in Sweden Internet Connection,’’ is typical for Instrumental Users,
(Nagelkerke R-squared = 31.6%), Instrumental Users in although this is not the case for Austria where Instrumental
Norway (Nagelkerke R-squared =17.3%), and Advanced Users tend to have broadband.
Users was best explained by the model in UK (Nagelkerke
R-squared = 13.7%). 4.4.5. Advanced Users
As described in Table 6, the predictor variables account
4.4.1. Non-User for more of the variance in the dependent variables for the
As shown in Table 6, access at home is important in all UK (13%) and Spain (10%) than for the other countries
countries, particularly in the UK, though it is relatively less (Nagelkerke R-squared). In Norway and Sweden, access is
important in Spain. The overall effect of age is also very the only significant predictor, although this is not a
high. Because access varies to a great extent between significant predictor for Austria. The roles of age and
countries, it is important to take it into account when gender are most important in Spain and the UK, but they
comparing the countries. Sweden and Norway have the are not important in the other countries.
132 P.B. Brandtzæg et al. / Int. J. Human-Computer Studies 69 (2011) 123–138
Table 6
A description of the results (Exp(B) and Nagelkerke R-square) from the regression analysis within in each country in the sample with the different user types
as dependent variables.
Variable Country
Non-Users as dependent
Gender 1.551 0.958 1.717nn 1.398nnn 1.235n
Age 2.565nnn 1.896nnn 2.250nnn 1.994nnn 2.640nnn
Household 1.498n 1.148 1.353nn 1.446nnn 1.587nnn
Access 0.264nnn 0.247nnn 0.283nnn 0.209nnn 0.333nnn
Nagelkerke R-squared 0.607 0.454 0.518 0.589 0.552
Note: N=12,666, sub-samples; Norway (N=451), Sweden (N =908), Austria (N=814), Spain (N=4424), UK (N=6069).
n
po 0.05.
nn
po 0.01.
nnn
po0.001.
Percentage of households who have broadband access be added to the entertainment activities in which the user is
at home already engaged.
80
Spain However, people who start to use the Internet as adults
70 Austria
will mostly do so for instrumental reasons; they use the
technology because it is useful, not because it is entertaining.
60 Sweden
There is reason to believe that adults starting to use the
United Kingdom technology first go through a period of being Sporadic
50
Norway Users, trying out the possibilities, and then ‘‘convert’’ to
40 being Instrumental Users to cope with practical tasks that
can be solved with the use of Internet services.
30 As the Internet evolves, the level of access and usage of
each of these groups in the user typology tends to increase,
20
but the divide in terms of different user types will probably
10 grow. Each group is progressing, but the most favored user
types, Advanced Users, Instrumental Users, and Entertain-
0 ment Users, might progress faster than the others, compared
2003 2004 2005 2006 2007 2008 with Non-Users and Sporadic Users. This is already a
Year pattern that has been found in both Norway and UK. The
more sophisticated the user type, the more we can expect the
Fig. 3. Percentage of households who have broadband access from 2003 to
2008 at home in countries included in this study (data from Eurostat). users to have the high degree of online competence and
knowledge that are needed to develop and adapt faster in
connection with the increasing development and distribu-
tion of the Internet. This anticipation of an increasing divide
The variation between countries may be dependent on between groups or new media user types is also illustrated in
education level, socioeconomic indicators, or national Evans and Wurster’s Blown to Bits (2000). This notion also
wealth in terms of GDP, as well as the infrastructure related parallels the ‘‘rich get richer’’ effect, where people who are
to the Internet and broadband in the specified country. active online benefit most from the Internet and will achieve
Moreover, as suggested by Livingstone et al. (2001), cross- new skills faster along with the development of new online
country research on media use has to take into account the services. This thought is also supported by the fact that users
context of the available media and the policies that regulate in Norway, where access and usage is high, have a more
them, in addition to a wide range of cultural factors that differentiated use pattern compared to users in other and
frame everyday life, such as differences in family structure less developed Internet countries in this study.
and linguistic uniformity/diversity. The same can be claimed
for online use. Future studies should therefore investigate a 5.2. Validity
larger number of socioeconomic indicators.
A possible threat to the validity of our results is that the
5.1. Future predictions—towards an increasing digital divide respondents’ Internet usage was measured using a self-
or user type divide? assessment approach, i.e., the respondents simply indicated
whether they were able to carry out a specific task related to
It is to be expected that the penetration and diffusion of computer use without their skills being assessed, tested, or
broadband, digitalization, and media convergence will actually observed. As a result, people might have wrongly
increase in the near future. It is reasonable to believe that, reported their skills. In any case, we believe that this threat is
together with the emphasis on new technology in schools, limited because several studies have indicated that people’s
this convergence will result in a trend towards more wide- own perceptions are a good indicator of their actual abilities
spread use of the Internet among younger generations. (Demunter, 2006).
Therefore, we expect the mean age of Entertainment Users In this context, the Internet user typology approach is
to decrease in the future. merely a convenient way of describing a heterogeneous set of
Entertainment Users are already the youngest group in data. As seen in the previous discussion, there is no claim
our sample, and we expect that the young users will convert that a user ‘‘is’’ a given user type for the rest of his/her life.
directly from being Sporadic to Entertainment Users as their There is also no claim that there are absolute boundaries
Internet skills grow. If all children have the opportunity to between the types; in the present approach, all users have a
use the technology, they will probably become Entertain- degree of belongingness to all the user types but are classified
ment Users after a short period of trying out all the according to the type that they most closely resemble.
possibilities. When they become older and want to broaden Another limitation is that the user types only cover Internet
their repertoire, they will become Advanced Users, meaning usage and not other media technologies. However, the
that in the future we will see fewer pure Instrumental Users. Internet must be viewed as one of the most important media
Instrumental activities of course will not disappear, but will channels in modern society.
P.B. Brandtzæg et al. / Int. J. Human-Computer Studies 69 (2011) 123–138 135
The selection of countries in this study is also crucial. Norway seem to be the most differentiated countries, with
Other countries have both higher (Iceland and the Nether- several prevalent user types, including both Non-Users and
lands) and lower coverage of Internet access (Czech a large number of Advanced Users.
Republic and Greece) than the countries in this study. On a general level, the majority of European citizens from
However, the countries selected in this study show some the countries investigated lag alarmingly behind in terms of
interesting variations in the relationships between user types Internet usage. 60% are either Non-Users (42%) or
and access. Sporadic Users (18%). The study also predicts the presence
Finally, when interpreting the results one should be aware of a ‘‘rich get richer’’ effect in which the divide between the
of the low statistical evidence in general. Most of the different user types will increase in connection with the
statistics in this study have low power. Readers should also growing development and distribution of the Internet and its
be aware that the use of secondary data involves several technologies. This development indicates larger inequalities
limitations, which are primarily related to a lack of flexibility of online participation in the future. The digital divide
in the measures included in this study. Nonetheless, this lack in terms of nurturing the more sophisticated user types
of flexibility is balanced by access to representative and should therefore remain a crucial issue on the agenda for
comparative cross-national survey data in countries of the technology developers as well as politicians, in the years
European Union. Even so, statistics on ICT use by house- to come.
holds may run into problems of international comparability The user type approach in this study extends the tradi-
because of structural differences in the composition of tional understanding of user acceptance models of the
households. Internet by not focusing on Internet acceptance as such but
rather on how adoption and use differ related to distinct
6. Conclusion and future work user types.
60
% within houshold category
Sporadic users
Instrumental users
50
Non users
Number of clusters Advanced users
40
Entertainment users
Fig. A1. The validity measure of Ray and Turi (1999) as a function of the 30
number of clusters (K) in the K-means analysis. Note: According to the
procedure shown in this figure, the validity measure is calculated as the ratio 20
between the mean intra-cluster distances, i.e., the distance between a point and
its cluster center, and the minimum inter-cluster distance; the minimum 10
distance between any two clusters. We calculated this measure for all solutions
for K=2–20. In general, the smaller the validity measure, the better the 0
separation between clusters. The two-cluster solution split the sample into users 1 2 3 4 or more
and Non-Users. The three-cluster solution divided the users into two groups: Members in houshold
high vs. low levels of Internet usage. Our judgment was that these distinctions
would not bring much new insight to the field, so we decided to extract five Fig. A4. User types by number of members in household (N=12,666).
clusters where the validity function had a clear local minimum. A pilot study Note: Non-Users decrease with the number of household members, while
also extracted five clusters. This final cluster solution has in addition theoretical Entertainment Users, and to some extent the Sporadic Users, increase. The
justification in Brandtzæg’s (2010) model of a unified media user typology. proportions of Instrumental and Advanced Users are relatively constant
across household category.
90
% withinn access category
50 80 Sporadic users
Sporadic users
% within gender category
45 70 Instrumental users
Instrumental users Non users
40 60
Non users Advanced users
35 50
Advanced users Entertainment users
30 40
Entertainment users
25 30
20
20
10
15
0
10 Access to PC no Narrowband Broadband Internet
No access Internet connection Internet connection connection
5
0 Access to PC and Internet at home
Male Female
Gender Fig. A5. User types per access category (N=12,666). Note: There is an
obvious decline in the proportion of Non-Users as access increases because
Fig. A2. User types by gender (N=12,666). Note: There are some the proportion of Advanced and Entertainment Users increases with
differences between the genders with regard to user types, but these are increased access. The most frequent user type among users that only have
not very large. There are generally more female than male Non-Users and access to a PC and not the Internet is the Sporadic Users, while the
somewhat more male Advanced and Entertainment Users. Instrumental Users typically have a narrowband connection.
P.B. Brandtzæg et al. / Int. J. Human-Computer Studies 69 (2011) 123–138 137
Table A1
Proportional sampling in the selected countries and the distribution of gender and population to mirror the real distributions.
Out of new sample (N) In new sample (N) Original sample (N) New sample (%) Population (%)
Table A2
Comparing the real distribution of population and gender in the selected countries with the original sample.
References Demunter, C., 2006. How Skilled are Europeans in Using Computers and
the Internet., Statistics on Focus. Industry, Trade and Services.
Barnes, S.J., 2007. Segmenting cyberspace: a customer typology for the Eurostat, Brussels.
Internet. European Journal of Marketing 41, 71–93. Dholakia, R.R., 2006. Gender and IT in the household: evolving patterns of
Bimber, B., 2000. Measuring the gender gap on the Internet. Social Science Internet use in the United States. The Information Society 22, 231–240.
Quarterly 81, 868–876. DiMaggio, P., Hargittai, E., Celeste, C., Shafer, S., 2004. Digital inequality:
Brandtzæg, P.B., 2010. Towards a unified Media-User Typology (MUT): a from unequal access to differentiated use. In: Neckerman, K. (Ed.),
meta-analysis and review of the research literature on media-user Social Inequality. Russell Sage, New York, pp. 355–400.
typologies. Computers in Human Behavior 26 (5), 940–956. Evans, P., Wurster, T., 2000. Blown to Bits: How the New Economics of
Brandtzæg, P.B., Følstad, A., Heim, J., 2003. Enjoyment. Lessons from Information Transforms Strategy. Harvard Business School Press.
Karasek. In: Blythe, M.A., Overbeeke, K., Monk, A.F., Wright, P.C. Hargittai, E., 2010. Digital Na(t)ives? Variation in Internet skills and uses among
(Eds.), Funology: From Usability to Enjoyment. Kluwer, Dordrecht, members of the ‘‘Net Generation’’. Sociological Inquiry 80 (1), 92–113.
the Netherlands, pp. 55–65. Hargittai, E., Walejko, G., 2008. The participation divide: content creation
Brandtzæg, P.B., Heim, J., Kaare, B.H., Endestad, T., Torgersen, L., and sharing in the digital age. Information, Communication and Society
2005. Gender differences and the digital divide in Norway—is 11 (2), 239–256.
there really a gendered divide? In: Childhoods: Children and Hargittai, E., Shafer, S., 2006. Differences in actual and perceived
Youth in Emerging and Transforming Societies, Gyldendal, Oslo, online skills: the role of gender. Social Science Quarterly 87,
Norway. 432–448.
Cohendet, P., 2003. The digital divide in the European enlarged economic Hargittai, E., 2002. Second-level digital divide: differences in people’s online
scenario: an assessment of the socio-economic effects. Report for ESA, skills. First Monday, 7,4. URL: /http://firstmonday.org/issues/
University Louis Pasteur, Strasbourg. issue7_4/hargittai/index.htmlS.
Colley, A., Maltby, J., 2008. Impact of the Internet on our lives: male and Heim, J., Brandtzaeg, P.B., Kaare, B.H., Endestad, T., Torgersen, L., 2007.
female personal perspectives. Computers in Human Behaviour 24 (5), Children’s usage of media technologies and psychosocial factors. New
2005–2013. Media & Society 9, 425–454.
Davis, F.D., 1989. Perceived usefulness, perceived ease of use, and user Horrigan, J.B., 2007. A Typology of Information and Communication
acceptance of information technologies. MIS Quarterly 13, 319–340. Technology Users. Pew Internet, Washington.
138 P.B. Brandtzæg et al. / Int. J. Human-Computer Studies 69 (2011) 123–138
Howard, P.E.N., Rainie, L., Jones, S., 2001. Days and nights on the Ortega Egea, J.M., Menéndez, M.R., González, M.V.R., 2007. Diffusion
Internet. the impact of a diffusing technology. American Behavioral and usage patterns of Internet services in the European Union.
Scientist 45, 382–404. Information Research 12 (2) Paper 302. URL: /http://informationr.
Johnson, M.G., Kulpa, A., 2007. Dimensions of online behavior. Toward a net/ir/12-2/paper302.htmlS.
user typology. Cyberpsychology & Behavior 10, 773–780. Peters, T., 2001. Spanning the digital divide: understanding and tackling the
Johnsson-Smaragdi, U., 2001. Media use styles among the young. In: issue. In: Bridges.org, Washington/Durbanville, p. 152.
Livingstone, S., Bovill, M. (Eds.), Children and their Changing Media Ray, S., Turi, R.H., 1999. Determination of number of clusters in K-means
Environment: A European Comparative Study. Lawrence Erlbaum clustering and application in colour image segmentation (invited paper).
Associates, Mahwah, NJ, London, pp. 131–141. In: Pal, N.R., De, A.K., Das, J. (Eds.), Proceedings of the Fourth
Karahasanović, A., Brandtzæg, P.B., Heim, J., Lüders, M., Vermeir, L., International Conference on Advances in Pattern Recognition and
Pierson, J., Lievens, B., Vanattenhoven, J., Jans, G., 2009. Co-creation Digital Techniques (ICAPRDT’99), Calcutta, India, 27–29 December.
and user-generated content—elderly people’s user requirements. Rogers, E.M., 2003. Diffusion of InnovationsFree Press, New York [1962].
Computers in Human Behavior 25 (3), 655–679. Selwyn, N., Gorard, S., Furlong, J., 2005. Whose Internet is it anyway?
Katz, E., Gurevitch, M., Hass, H., 1973. On the use of mass media for Exploring adults’ (non)use of the Internet in everyday life. European
important things. American Sociological Review 38 (2), 164–181. Journal of Communication 20, 5–26.
Kozinets, R., 1999. E-tribalized marketing?: the strategic implications of virtual Tan, P., Steinbach, M., Kumar, J., 2006. Introduction to Data Mining.
communities of consumption. European Management Journal 17, 252–264. Pearson, Addison Wesley, New York.
Livingstone, S., Helsper, E., 2007. Gradations in digital inclusion: children, Taylor, T.L., 2006. Play Between Worlds: Exploring Online Game
young people and the digital divide. New Media & Society 9, 671–696. CultureMIT Press, Cambridge, MA.
Livingstone, S., d’Hanes, L., Hasebrink, U., 2001. Researching young Van Dijck, J.D., 2009. Wikinomics and its discontents: a critical analysis
people and the changing media environment. In: Livingstone, S., Bovill, of Web 2.0 business manifestos. New Media & Society 11 (5),
M. (Eds.), Children and their Changing Media Environment: A European 855–874.
Comparative Study. Lawrence Erlbaum Associates, Mahwah, NJ, Venkatesh, V., Morris, M.G., David, G.B., Davis, F.D., 2003. User
London, pp. 31–51. acceptance of information technology: toward a unified view. MIS
Loosen, W., 2002. The Second-Level Digital Divide of the Web and its Quarterly 27 (3), 425–478.
Impact on Journalism. First Monday, 7,8. URL: /http://firstmonday. Webster, F., 2006. Theories of the Information Society. Routledge,
org/issues/issue7_8/loosen/index.htmlS. London.
OFCOM, 2008. Social Networking. A quantitative and Qualitative Wellman, B., Hogan, B., 2004. The Internet in everyday life. In: Bainbridge,
Research Report into Attitudes, Behaviours and Use. Office of W.S (Ed.), The Berkshire Encyclopedia of Human Computer Interac-
Communication, London. tion. Berkshire Publishing, Great Barrington, MA, pp. 389–397.