Transcript Digital Divide’ Reconsidered: A Country

‘Digital Divide’ Reconsidered:
A Country- and Individual-Level
Typology of digital inequality in 26
European Countries
Boris Kragelj, University of Ljubljana
[email protected]
Elmar Schlüter, University of Bielefeld
[email protected]
DIGITAL DIVIDE (definition)
First definition: difference between technology
„haves“ and technology „have not´s”
• Inequalities between individuals, households, companies
and regions regarding the access and use of new ICT
resources (specially internet):
• Inequalities in access to ICT (first level d. divide)
• Inequality in skills of using the internet (second level d. divide)
• Inequalities in ICT penetration between courtiers:
• Inequalities in access, use and skills to use ICT between more
and less developed countries (global digital divide)
... is empirical concept that refers to new form of
social inequality, regarding the access and/or use
of ICT and/or skills of using ICT (specially
internet) at various levels of society.
DIGITAL DIVIDE (existing studies)
• Studies on d. divide varies considering:
• Definition and level of d. divide under study (first level,
second level d. divide)
• Unit and societal level of study (individual, global)
• Dependent variables for measuring d. divide (internet access,
internet use, skill of internet use...)
• Independent variables for describing risk groups (gender, age,
education, income ....)
• Different methods of analysis applied or different indexes of
inequality developed (absolute dif., relative dif., time distance,
digital divide index...)
• Various approaches increased scholarly understanding of
the ‘Digital Divide’ in many ways, but they all study only
certain view of the whole Digital Divide phenomena even
though they might be strongly interrelated.
Aim of the present study
• Extending previous work on d. divide with integration of different
analytical perspectives on digital divide studies within one single
(multilevel latent class) model to examine their interrelations
• linking individual and societal level of analysis on both (first and
second) levels of digital divide to simultaneously provide for
individual and country level typology of groups regarding the
inequality in access, use and skills of internet use
• Explaining differences between these types of digital inequality
according to relevant independent socio-demographic and agregatelevel variables (on individual and societal level of analysis
respectively)
MODELING
Individual level of analysis
Societal level of analysis
First level dig.
divide
diff. in access and use of
internet between individuals
diff. in access and use of
internet between countries
Second level
dig. divide
diff. in skills of using internet diff. in skills of using
between individuals
internet between countries
Research questions:
• Which groups of digital inequality (regarding
internet access, use and skills) can be observed on
the individual level, and how this groups differ
across different (groups) of countries?
• What
are
individual
and
country-level
determinants for individual and country
affiliations to the (individual and societal) groups
of digital inequality?
Research model:
Multilevel latent class model
Country
level
Step 4
Gini coef.
% GDP for
ICT exp.
Step 3
Ccountry
Tel. Price.
Gender
Age
Step 2
Educat.
Income
Individual
level
Access
Cind
Int. use
Int. skills
Step 1
Data and variables
Individual level variables
Country level variables
n=21207
Internet access at home 1 with access
2 w/o access
Internet use
1 regular (last 4 weeks)
2 occasional (last 12 months)
3 non-users
internet skills
1,00 very confidant
2,00 less confidant
3,00 non internet user
%
36,3
63,7
40,4
7,4
52,2
28,0
19,8
52,2
gender
48,1
51,9
16,2
44,6
21,4
17,8
19,4
25,1
25,1
30,4
32,1
33,2
22,1
12,6
age group
household income
education (classes)
1 male
2 female
1 up to 24
2 25 to 49
3 50 to 64
4 65 and more
1 low
2 avarage-low
3 avarage-high
4 high
1 low
2 medium
3 high
4 still studying
Source: SIBIS and SIBIS+
Statistical indicators benchamrking the information
society (EU/FP5 project)
n=26
% of GDP for ICT expenditure 1
2
3
GINI coeficient
1
2
3
telecom prices
1
2
3
low
medium
high
low
medium
high
low
medium
high
%
36,4
29,1
34,5
41,2
29,5
29,3
39,5
34
26,5
Source: EUROSTAT – structural indicators
Country (sample) statistics
county
n
weighted avarage
n
weight
country
n
weighted avarage
n
weight
austria
500
214
0,43
sweden
500
237
0,47
belgium
585
322
0,55
switzerland
522
201
0,38
denmark
501
143
0,29
united
1000
3138
3,14
finland
669
185
0,28
bulgaria
1008
427
0,42
france
1000
3124
3,12
czech
1096
598
0,55
1001
73
0,07
1000
544
0,54
germany
1001
4407
4,40
estonia
greece
505
296
0,59
hungary
ireland
500
105
0,21
latvia
994
126
0,13
italy
1000
3048
3,05
lithuania
1017
189
0,19
luxemburg
500
12
0,02
poland
1000
2041
2,04
neterlands
530
456
0,86
romania
1054
1220
1,16
portugal
500
271
0,54
slovakia
1199
344
0,29
spain
1015
2213
2,18
slovenia
1002
105
0,10
Total
21207
24037
26
Results: individual level (3) latent class solution
fit statistics for exploratory analysis of latent class
models (individual level)
LL
1-Cluster
2-Cluster
3-Cluster
4-Cluster
5-Cluster
-61689
-40013
-39062
-39061
-39061
BIC(LL) AIC(LL) Npar df p-value
123428
80117
78254
78294
78335
123388
80045
78149
78157
78165
1,0
0,8
0,6
0,4
0,2
C luster1
C luster2
C luster3
ccd_rec
0-1 Mean
inet_u
0-1 Mean
inet_ah
0-1 Mean
0,0
5 12 1,5e-9808
9 8 1,6e-406
13 4
0,560
17 0
.
21 -4
.
relative size of the clusters and
conditional probabilities in relation to
digital divide variables (3 cluster
solution)
Cluster1 Cluster2 Cluster3
Cluster Size
0,52
0,34
0,14
Indicators
Inet. access
with access
w/o access
Mean
Inet use
regular
occasional
non-users
Mean
Int. skills
very confidant
less confidant
non int. user
Mean
0,08
0,92
1,92
0,82
0,18
1,18
0,31
0,69
1,69
0,00
0,00
1,00
3,00
0,99
0,01
0,00
1,01
0,50
0,50
0,00
1,50
0,00
0,00
1,00
3,00
0,71
0,29
0,00
1,29
0,27
0,73
0,00
1,73
individual level (3) latent class solution with covariates
Qualitative characteristics of latent classes and their
inequality according to relevant independent socio
demographic variables
test of significance and conditional probabilities for
covariates (effects on membership in latent classes)
Covariates
gender
male
female
age
up to 24
25 to 49
50 to 64
65 and more
hous. income
low
avarage-low
avarage-high
high
education
low
medium
high
still studying
Cluster1 Cluster2 Cluster3 Wald p-value
0,44
0,56
0,05
0,38
0,26
0,30
0,58
0,42
0,25
0,59
0,14
0,02
158,9 3,0E-35
name
1371,3
Class1:
Non
users
0,48
0,52
0,18
0,57
0,20
0,05
1272,4
0,30
0,31
0,21
0,18
0,07
0,15
0,28
0,51
0,14
0,28
0,37
0,22
0,00
0,10
0,25
0,34
0,31
1420,6
0,46
0,38
0,14
0,02
0,00
0,22
0,38
0,29
0,11
0,00
description
individuals without
internet access and
skills of using internet
which don't use the
internet
Class2:
individuals with internet
Heavy
access at home and
users
good internet skills,
which use the internet
on daily basis
Class3:
individuals without
Out of home
internet access at
moderate
home, with moderate
users
internet skills, who use
internet occasionally
(out of home)
demographic
characteristics
High probability on being
older, female, having low
income and low education
High probability on being
young, male, and having
high income and high
education
Higher probability on
being middle age, female,
with average income and
average education.
group level (5) latent G-class solution
fit statistics for exploratory analysis of group
level latent class models (country level)
LL
3-Cluster 1-GClass
*3-Cluster 2-GClass
*3-Cluster 3-GClass
*3-Cluster 4-GClass
3-Cluster 5-GClass
3-Cluster 6-GClass
3-Cluster 7-GClass
-37880
-37133
-37071
-36667
-36044
-36064
-36044
relative size of G-classes and conditional probabilities in
relation to membership in the individual level classes (5 G-class
solution)
BIC(LL) AIC(LL) Npar
75892
74428
74335
73556
72341
72410
72402
75786
74298
74181
73378
72139
72183
72151
13
16
19
22
25
28
31
GClass1
0,38
GClass Size
Clusters
non users
heavy users
moderate users
0,57
0,28
0,15
GClass2 GClass3 GClass4 GClass5
0,30
0,22
0,08
0,01
0,76
0,05
0,19
0,31
0,55
0,15
0,41
0,44
0,15
0,34
0,28
0,38
Distribution of countries across G-classes
0,80
GClass1 GClass2 GClass3 GClass4 GClass5
0,70
france
italy
portugal
spain
czech
slovenia
0,60
0,50
0,40
0,30
0,20
0,10
0,00
non users
GClass1
heavy users
GClass2
GClass3
out of home moderate users
GClass4
GClass5
greece
bulgaria
hungary
latvia
lithuania
poland
romania
slovakia
denmark austria
estonia
germany belgium
ireland
finland
neterlands luxemburg
sweden
switzerland
united
(5) latent G-class solution with G-level covariates
test of significance and effects of group level covariates on
group class level membership (effect coding)
Covariates
GClass1
GClass2
GClass3
telecom prices
low
2,2
-3,9
1,1
medium
-0,7
-0,1
0,0
high
-1,5
3,9
-1,2
gini coeficient
low
1,7
-3,0
-0,5
medium
-0,7
0,7
1,5
high
-1,0
2,3
-1,0
% of GDP for telecom. expenditure
low
0,7
1,6
-0,3
medium
0,9
-3,2
0,2
high
-1,6
1,6
0,1
GClass4
3,1
1,1
-4,2
GClass5 Wald p-value
24,3
0,0
22,4
0,0
19,8
0,0
-2,5
-0,4
2,9
3,8
0,0
-3,8
-2,0
-1,6
3,6
-2,9
2,1
0,9
1,0
0,0
-1,0
Qualitative characteristics of group level latent classes and their inequality according to relevant
structural covariates
name
description
Structural characteristics
GClass1: More than half are non users, one third Low telecom prices and low gini coefficient increase
Europan heavy users and small group of moderate the odds of membership, high telecommunication
avarage
users
expenditures as share of GDP decreases the odds
GClass2:
High telecom prices and high gini coefficient
Laging
A majority are non users, small group of
increase the odds of membership, medium tel.
moderate users, almost no heavy users
behing
expenditure decreases the odds
Gclass3:
Leaders
Low telecom prices increases, low telecom.
expenditure decreases and medium gini coefficient
increases odds of class membership
Gclass4:
Low telecom prices and low gini coefficient increase
Almost half of heavy users, less than half
Above
the odds of class membership, low telecom
avarege
non-users
expenditures decreases the odds of class
membership
Class5:
Predominating moderate, out of home
High telecom. prices and high gini coefficient and
Moderat
users with almost the same share across low telecom expenditures increase the odds of class
e equity
all individual classes
membership
More than half are heavy users, only one
third of non users
Summary & Conclusion
Using standard digital divide variables (access, use, skills of internet use) within “multilevel
latent class framework” we have identified three types of classes regarding digital
(in)equality on individual level, whose distribution differ significantly among five group of
EU countries:
On ind. level we identified classes of (1) heavy users and (2) out of home moderate users (this
group is different than anticipated!!!), and (3) non-users, the last group is presenting dig.
divide risk group (52% of population) mainly represented by female, older, low income and
low education individuals.
On soc. level we identified five groups of countries: (1) ICT leaders, (2) above ICT average
(both mainly north of Europe), (3) ICT average (south of Europe), (4) lagging behind
(eastern European block + Greece) and special one country (EE) class, with high share of out
of home users and equal distribution between all individual level classes. These groups of
countries varies considerably regarding structural characteristics such as ICT expenditure as
% of GDP, gini coefficient and telecom prices
Risk group of countries that are lagging behind (presenting 1/3 of European countries) are
characterised by high telecom prices and high income inequality, showing that much of
digital divide in Europe can be explained with traditional forms of social inequality and poor
national telecomm. policies (blocking free market, low investment in ICT...)
Digital divide as a new form social inequality is not new at all. Once again it only reflects
already well established traditional forms of social inequality in Europe through a new
perspective: risk groups of individual and countries are staying the same.
Methodological Discussion
We found ML-LC to be useful when simultaneously searching for typologies and groups on
two different levels of analysis, and at the same time explaining and predicting these
group membership with other relevant (individual and structural) variables ... Still (as the
method is not so well documented) we run into several problems:
 non convergence: some multilevel models with less DF converge better than one with
more DF! How to explain this? Are results valid at all?
 Choosing the best model solution:
- Why at some point (when you move from individual level to multilevel framework,
or when you introduce covariates) the model fit is considerably worse than before,
even though you catch more variance and reduce the classification error?
- How can you test if one model fits significantly better that the other in multilevel
framework where you don’t have chi square statistics to conduct chi square test?
- Should one search for best multi level solution before or after introducing covariates
on individual level, and with or without covariates on structural level?
 Choosing non-parametric vs. parametric multilevel model: In our case we have modelled
existence of groups of countries on aggregate level (non parametric ML-LC???), but our
solution showed that these groups actually (almost) show an order from leaders to
laggards in terms of ICT! Would it be better to model a continuum (factor) of countries on
the aggregate level (parametric ML LC, and how to test for this?
Thank you for attention!
‘Digital Divide’ Reconsidered:
A Country- and Individual-Level Typology of digital
inequality in 26 European Countries
Boris Kragelj, University of Ljubljana
[email protected]
Elmar Schlüter, University of Bielefeld
[email protected]
group level (5) latent G-class solution with macro level
covariates
test of significance and effects of group level covariates on
group class level membership
Covariates GClass1
GClass2 GClass3 GClass4 GClass5
telecom prices
low
10,8
-0,4
medium
2,8
-0,6
high
0,0
0,0
gini coeficient
low
9,2
0,1
medium
6,4
1,5
high
0,0
0,0
% of GDP for telecom. expenditure
low
1,5
0,1
medium
-0,1
-3,5
high
0,0
0,0
17,6
11,7
0,0
10,4
11,1
0,0
-5,8
-6,3
0,0
6,8
-1,7
0,0
12,4
10,7
0,0
1,7
5,4
0,0
Wald p-value
32,2
0,0
19,4
0,0
46,8
0,0
0
0
0
0
0
0
0
0
0