Content uploaded by Josef (Kuo-Hsun) Ma
Author content
All content in this area was uploaded by Josef (Kuo-Hsun) Ma on Sep 13, 2021
Content may be subject to copyright.
Vol.:(0123456789)
Social Indicators Research (2019) 144:133–166
https://doi.org/10.1007/s11205-018-2030-0
1 3
National Income, Political Freedom, andInvestments inR&D
andEducation: AComparative Analysis oftheSecond Digital
Divide Among 15‑Year‑Old Students
JosefKuo‑HsunMa1 · ToddE.Vachon2· SimonCheng3
Accepted: 9 November 2018 / Published online: 17 November 2018
© Springer Nature B.V. 2018
Abstract
Digital technology has become an indispensable component in education around the world.
Despite its growing importance, a gap in students’ digital skills and usage based on their
socioeconomic status—known as the second digital divide—has been identified in a wide
range of countries. Using data from the 2009 OECD Programme for International Student
Assessment, we consider two aspects of the second digital divide for 15-year-olds across
55 countries: the gaps in use of educational software at home and Internet literacy. Spe-
cifically, we ask whether national income, political freedom, and national investments in
research and development (R&D) and secondary education are associated with the second
digital divide. We find that national income predicts the digital divide and that national
investments have differential effects depending upon a country’s income. R&D spending
reduces the socioeconomic gap in educational software use only in low-income countries.
Educational expenditures reduce the Internet literacy gap in high-income countries while
exacerbating it in low-income ones. Additional analyses suggest that income inequality
increases the digital divide, but like political freedom, the effects become non-significant
when national income is considered. We conclude by discussing the implications of these
findings for policymakers interested in reducing the digital divide.
Keywords Second digital divide· R&D· Educational expenditures· PISA· Multilevel
modeling
* Josef Kuo-Hsun Ma
kuo-hsun.ma@uconn.edu
Todd E. Vachon
todd.vachon@rutgers.edu
Simon Cheng
simon.cheng@uconn.edu
1 Department ofSociology, National Taipei University, Taipei, Taiwan
2 Department ofLabor Studies andEmployment Relations, Rutgers, The State University ofNew
Jersey, NewBrunswick, NJ, USA
3 Department ofSociology, University ofConnecticut, Storrs, CT, USA
134
J.K.-H.Ma et al.
1 3
1 Introduction
In recent decades, the rate at which digital technology has penetrated educational settings
throughout the world has been staggering. Familiarity with hardware such as desktop and
tablet computers, and software such as word-processing programs and Internet search
engines is rapidly becoming a prerequisite to success. However, despite the progressive
spread of new technology, a number of inequalities have been identified. The first dispar-
ity, referred to as “the first digital divide,” concerns the socioeconomic inequality between
people who have access to digital technology, with more affluent people having greater
digital access. Recent studies have found a worldwide decline in the digital access divide,
even within less-developed countries. This is largely due to the intentional efforts of educa-
tors, policymakers, and entrepreneurs (Erichsen and Salajan 2014; Spring 2008). The sec-
ond disparity, known as the “second digital divide”, concerns the socioeconomic disparity
in digital use for learning and productive purposes (Attewell 2001). In sum, students with
higher socioeconomic status are more likely to use computers for educational purposes.
This inequality exists even after digital access equality has been achieved, and it remains a
problem for both high- (Hargittai and Hinnant 2008; Peter and Valkenburg 2006) and low-
income countries (Drori 2010; ITU 2011). This raises the question: Why does the second
digital divide persist across countries of different economic standing, especially affluent
nations that have largely eliminated the access divide?
Understanding the cause of the second digital divide, as well as identifying possible
solutions to it, is increasingly relevant as digital technology becomes further ingrained into
global education systems. Schools are increasingly posting assignments online and requir-
ing students to use computers for schoolwork (OECD 2011). Lacking parental guidance,
students of lower socioeconomic status (SES) are more likely to use digital devices for
social networking and gaming and thus are less prepared for academic assignments than
their higher-SES peers, whose parents tend to guide their children in using digital technol-
ogy for productive purposes (Attewell 2003; van Deursen and van Dijk 2014). This soci-
oeconomic gap in digital use may reinforce or even exacerbate existing educational and
social inequalities. Earlier research suggests that several demographic features, such as race
and family structure, are related to the digital divide (DeBell and Chapman 2006; Notten
etal. 2009), but SES remains the most influential individual-level factor in cross-national
research on digital use (OECD 2011, 2015).
Moreover, research highlighting school-level explanations suggests that low-SES stu-
dents tend to use computers for remedial purposes, whereas their high-SES peers are more
likely to enroll in advanced computer classes and use technology for innovative purposes
(Becker 2000; Robinson 2014). In the U.S., students who attend schools in economically
advantaged neighborhoods are 1year ahead of those attending less privileged schools when
it comes to online research and comprehension skills (Leu etal. 2014). This again shows
that the digital divide by SES remains persistent when schools fail to provide sufficient
computer-related resources and courses to help develop the digital skills of underprivileged
students (Natriello 2001).
The relationship between SES and digital use may also depend on national context. Pre-
vious research has identified several country-level factors that are associated with the first
digital divide, including economic development (Guillén and Suárez 2005; Norris 2001;
Robison and Crenshaw 2010), digital infrastructure development (Dutton etal. 2004), and
the degree of democracy in government (Corrales and Westhoff 2006; Robison and Cren-
shaw 2010). However, little empirical research has examined how country-level contextual
135
National Income, Political Freedom, andInvestments inR&D…
1 3
factors may moderate the second digital divide, including how various nation-level invest-
ments might affect the relationship between family SES and digital use.
To begin to address these gaps in the literature, we examine the second digital divide
among 15-year-old students in 55 countries, using data from the 2009 Programme for
International Student Assessment (PISA) survey. We define the second digital divide as the
effect of socioeconomic status on two proxy measures of digital use—inequality in com-
puter use for educational purposes at home and Internet literacy. Broadly, we ask how insti-
tutional arrangements of countries, such as macro-economic conditions, public policies,
and business practices, affect the level of the second digital divide. Following previous
research on digital inequality, we focus first on the role of national income and political
freedom in explaining cross-national variation in the second digital divide. Because our
study focuses on youths, we go on to explore two types of national-level investments that
have received much attention in comparative educational research (Dale 2005; Erichsen
and Salajan 2014; Spring 2008) and studies of global digital inequality (Drori 2006, 2010;
Dutton etal. 2004; Norris 2001) and may be related to the second digital divide among
15-year-olds: investment in research and development (R&D) and national expenditures on
secondary education.
To date, comparative education and stratification research has focused primarily on
achievement gaps, but students’ digital skills and literacy are an important aspect of ine-
quality in the digital era. This paper builds on previous research and makes several con-
tributions to the literature. First, past studies have focused on the first digital divide for
adults. We examine the second digital divide among 15-year-old students—a section of
the population for which the digital divide may prove most important. Second, previous
research has focused primarily on wealthy countries, especially the U.S. (DiMaggio etal.
2004). We go beyond these studies by examining an economically diverse set of countries.
Finally, the existing research has focused on individual-level explanations for the digital
divide (Robison and Crenshaw 2010). We consider how the influence of family SES—an
important individual-level factor—varies cross-nationally. Taken together, we believe this
research provides insight into some of the key factors associated with the second digital
divide, which may prove valuable for countries seeking to become successful players in the
global knowledge economy and to increase the digital literacy of their populations (Spring
2008).
2 Individual‑Level Factors Associated withtheDigital Divide
Economic globalization and the emergence of the information society have dramatically
increased competition among individuals seeking access to a limited number of opportuni-
ties. However, middle-class and elite families have been able to mobilize various resources,
including digital learning resources, “to stand out from the crowd” (Brown 2013, p. 683).
This process of positional competition has increased the importance of digital technology
for individuals seeking to increase their knowledge and establish social networks that can
foster social mobility (DiMaggio etal. 2004; DiMaggio and Cohen 2005). If the distribu-
tion of technology is unequal, it may only exacerbate already existing inequalities. Earlier
research suggests that students from racial-minority or single-parent families are less likely
to access the Internet at home than their non-minority peers or those living in two-parent
households (DiMaggio et al. 2004; Notten et al. 2009). However, when socioeconomic
background is controlled for, the effects of race and family type are reduced significantly
136
J.K.-H.Ma et al.
1 3
(DeBell and Chapman 2006; OECD 2011; Peter and Valkenburg 2006). In other words,
SES has been found to be the most influential predictor of the digital divide (OECD 2015).1
Previous research offers several clues as to why SES is an important factor when study-
ing the digital divide. For example, recent studies from the United States (Hargittai and
Hinnant 2008) and Switzerland (Bonfadelli 2002) suggest that highly educated adults
are more likely to do capital enhancing or information-oriented activities online, such as
visiting websites about national news, health, and financial information, than their less-
educated counterparts. In Korea, higher-SES adults are more likely to use the Internet for
political knowledge (Kim 2008). This disparity may be due, in part, to the conscious deci-
sions made by higher-SES adults who are seeking to improve or maintain their social sta-
tus, but high SES is also related to higher levels of education in general, which can mediate
the relationship between SES and digital use.
While much of the previous literature has focused on adults, the digital divide among
school-aged children and adolescents is equally, or even more, important. The way that
students use digital technology can influence their learning as well as their non-academic
behavior (see Attewell etal. 2003; Fuchs and Wossmann 2004), and the digital inequalities
generated during youth are likely to be carried into adulthood. For example, research from
Britain (Livingstone and Helsper 2008) and Australia (Smith etal. 2013) has shown that
students with highly educated parents are more likely to use computers or the Internet for
learning. By comparison, in the Netherlands (Peter and Valkenburg 2006) and Hong Kong
(Leung and Lee 2012), lower-SES students use the Internet primarily for gaming or social
networking. In part, this can be attributed to parents’ investment choices when purchasing
technology (e.g., digital devices with more functionality, useful software, and high-speed
Internet) as well as their active involvement in their children’s use of the technology, such
as sitting with and supervising their children’s computer use (Attewell 2001).
3 Country‑Level Factors Associated withtheDigital Divide
In the final decades of the 20th century, there were just a few national-level efforts to pro-
mote the use of technology in education (Erichsen and Salajan 2014; U.S. Department of
Education 1996). Since then, the quantity and urgency of these efforts has increased dra-
matically (Spring 2008), often inspiring competition among the most developed nations
(Erichsen and Salajan 2014). Studies in less-developed countries indicate a growing com-
mitment to increasing digital literacy as well, but their efforts are often limited by the
highly unequal global distribution of resources (Drori 2006, 2010; UNESCO 2015). To
further examine how this global inequality affects a country’s ability to bridge the digital
divide among students, we first consider how national income and political freedom—two
strong predictors of the first digital divide—may influence the second digital divide. Then,
we focus on national investments in R&D and secondary education because they may be
relevant for the youth population examined in this study. Furthermore, unlike political free-
dom and national income, which are resilient features of economies, national investments
1 Research from Australia (OECD 2011) and Cyprus (Milioni et al. 2014) suggests a “reverse digital
divide” between immigrants and nonimmigrants; that is, there is a “compensatory or remedial use of the
Internet (Milioni et al. 2014, p. 333)” by racial minority immigrants in order to overcome their existing
racial and social barriers.
137
National Income, Political Freedom, andInvestments inR&D…
1 3
can fluctuate more rapidly and thus may be capable of affecting digital inequality in the
near term.
3.1 National Income
National income is believed to be one of the most influential factors in Internet diffusion
(Guillén and Suárez 2005; Norris 2001), which is supported by the persistently low rates of
computer ownership and Internet access in poor countries (Drori 2006; ITU 2011). With-
out widespread diffusion of digital technology, access is generally restricted to the higher-
SES members of society. In this way, national income can moderate the level of the digital
divide by reducing the disparity in access between the more and less affluent.
As economic development improves people’s living conditions, low-SES people spend a
smaller proportion of their income on necessities (e.g., food, housing) and thus have more
money to spare for digital devices. Moreover, economic growth can promote both public
expenditures on and private investments in Internet infrastructure (e.g., high-speed Inter-
net landlines and community e-service) and stimulate competition among Internet service
providers, which can reduce the price charged for access (Hilbert 2010). Whether due to
increases in disposable income, improved digital infrastructure, or increased competition,
economic development has been shown to reduce the first digital divide by improving
Internet access rates for those on the low end of the SES spectrum. While there are no
direct comparative studies of the second digital divide among youths, we expect a simi-
lar relationship between national income and the gap in digital use. That is, increases in
national income should promote the use of digital technology for learning and productive
purposes among low-SES students, leading to reductions in the level of the second digital
divide:
Hypothesis 1 National income will have a negative association with the second digital
divide; that is, increased national income will reduce the socioeconomic gap in digital use.
3.2 Political Freedom
Distinct from national income, political freedom may also influence the digital divide
(Norris 2001). In less-democratic or authoritarian societies, the government may prevent
the diffusion of online networked technology by monitoring people’s online activities and
blocking Internet access in public spaces such as libraries or schools. The fear of govern-
ment reprimand and the lack of digital technology at home are likely to work together to
reduce digital use by the less-privileged in authoritarian societies. In contrast, basic civil
liberties such as freedom of speech and press are protected in democratic states, opening
the door to more public investments in digital access points at schools and libraries. Fur-
thermore, political freedom can also spur the growth of private investment in new informa-
tion technologies, which may reduce the cost of Internet access, and make it more readily
available to the less affluent (Robison and Crenshaw 2010).
The intuitiveness of this argument notwithstanding, some scholars suggest that eco-
nomic factors tend to outweigh political freedom when it comes to the digital divide. For
example, Corrales and Westhoff (2006) explain that not all authoritarian governments
are against Internet diffusion. In particular, authoritarian states that are market oriented
have fewer restrictions on Internet use. In line with this argument, Robison and Crenshaw
(2010) demonstrate that democratic governance and political stability only affect Internet
138
J.K.-H.Ma et al.
1 3
development when a country’s economy is depressed or stagnant. Again, these studies on
political freedom have focused only on the first digital divide, with little attention to the
second, but we see no reason to anticipate that the findings would be different. Thus, based
on these accounts, we propose that:
Hypothesis 2 Political freedom will have a negative association with the second digital
divide among students; that is, increased political freedom will reduce the socioeconomic
gap in digital use.
3.3 R&D
Many scholars have referred to the rapidly changing world economy as the new “knowl-
edge economy,” characterized by massive information flows, extensive adoption of new
information technology, and rapid advances in science and technology (Dale 2005; Powell
and Snellman 2004; Spring 2008). Spending on R&D, by both the public and the private
sectors, is indicative of a country’s efforts to become more competitive and innovative and
to play an active role in the knowledge economy. However, the benefits of national-level
investments in R&D are not limited just to the economy. R&D produces spillover effects
that can affect many other areas of social life, including agriculture, medicine, entertain-
ment, and, most relevant to our study, education (Drori 2006).
R&D spending is often piloted in educational settings with the aim of improving edu-
cational outcomes of disadvantaged students and enhancing the overall quality of educa-
tion (Snow 2002; Spring 2008). Heyneman and Loxley (1983, pp. 1183–1184) note that
“the areas of the world with comparatively large amounts of research and development
capital tend also to be the areas where educational paradigms are invented.” For example,
the Australian government created a platform that enables schools to share and distribute
educational resources through online portals (Education Services Australia 2012). In the
U.S., the invention of cyber schools is regarded as a new opportunity for students who
have failed in the conventional school system (Hill 2010). Additionally, major corpora-
tions throughout the world, including Microsoft, Intel, Hewlett-Packard, and AT&T, have
donated a great deal of computer equipment to underserved schools (Norris 2001). For
less-developed countries, Information and Communication Technologies for Development
(ICT4D) is a global initiative that works closely with governments, universities, public
schools, and private organizations to reduce digital inequality (UNESCO 2016). Consider-
ing these findings, we offer the following hypothesis:
Hypothesis 3 National investments in R&D will have a negative association with the
second digital divide; that is, increased investment will reduce the socioeconomic gap in
digital use.
3.4 Educational Expenditures
Globally, the level of spending per pupil on primary and secondary education has surged
over the past two decades (Baker etal. 2002), marking a substantial increase in national
investments in the production of human capital. This investment affects the educational
outcomes of students, particularly those from less affluent households (Chiu 2010). As a
result, many scholars have explored the relationship between public education expenditures
139
National Income, Political Freedom, andInvestments inR&D…
1 3
and the educational attainment gap between low- and high-income children (for results
from the U.S., see Mayer 2001; for results based on international comparisons, see Vegas
and Coffin 2015).
Public expenditures in educational settings may level the playing field for students
across socioeconomic backgrounds and subsequently reduce the second digital divide.
Qualitative studies suggest that insufficient digital access and lack of guidance in the use
of digital technology for less-affluent students contributes to substantial levels of inequality
in digital use (Goode 2010; Natriello 2001; Robinson 2014). For instance, Natriello (2001)
points out that racial minorities and socioeconomically underprivileged students in the
U.S. are more likely to attend schools with extremely low educational quality and severe
budget deficits that contribute to their deficiency in digital skills.
This suggests that the digital divide is, in part, due to the unequal distribution of edu-
cational resources among schools, including the availability of computer programs. From
this line of reasoning, it is logical to suppose that government investments in digital access
and the creation of opportunities for disadvantaged students to develop their digital skills
would lead to a reduction in the second digital divide (Gamoran 2001). Considering the
national variation in public spending on education, even among countries with similar eco-
nomic standing, we propose the following hypothesis:
Hypothesis 4 Investments in secondary education will have a negative association with
the second digital divide; that is, increased investment will reduce the socioeconomic gap
in digital use.
3.5 Dierential Eects byNational Income Level
In studying the second digital divide, it is important to account for the vast diversity of life
experiences and educational trajectories in different regions of the world, especially noting
the institutional variation between developed and less-developed countries (Buchmann and
Hannum 2001; Juárez and Gayet 2014). Since we are focusing on school-aged youths, we
note that national investments in human capital may be more effective at reducing the edu-
cational achievement gap in high-income countries that already provide basic educational
resources, including learning materials and teacher training (Vegas and Coffin 2015). Con-
versely, the high levels of social inequality and poverty that exist in low-income countries
can limit the benefits of educational investments to only the most affluent students (Juárez
and Gayet 2014). For example, recent research in Moldova suggests that poverty and lack
of employment opportunities can affect how children learn at home and in school. There is
also a shortage of qualified teachers, as many teachers emigrate to other countries for work
(Worden 2014). In addition, the inability of many poor countries’ governments to provide
education leads to an increased reliance on private schools, which has greatly limited edu-
cational opportunities for poor and underprivileged students (Chankseliani 2014).
Conversely, the marginal utility of further human capital investments in high-income
countries may be diminished because of already high levels of investment (Buchmann and
Hannum 2001). The persistence of educational inequalities in affluent countries, despite
their high levels of investment in education, offers some support for this argument (Han-
num and Buchmann 2005; Raftery and Hout 1993). Considering these possibilities, we
explore how the association between national investments in R&D and secondary educa-
tion and the digital divide may vary depending upon the national income of a country. We
offer the following two competing hypotheses:
140
J.K.-H.Ma et al.
1 3
Hypothesis 5a The effects for R&D spending and educational expenditures in bridg-
ing the socioeconomic divide in digital use will be smaller for students in low-income
countries.
Hypothesis 5b The effects for R&D spending and educational expenditures in bridg-
ing the socioeconomic divide in digital use will be greater for students in low-income
countries.
We explore these hypotheses by creating interaction effects between our two national-
level investment variables—R&D spending and educational expenditures—and the
national-level measure of income. These interaction effects will reveal any variations in the
findings of our main models that could be attributed to levels of economic development.
We see no theoretical reason to pursue similar interactions for political freedom, but results
from such models are available upon request.
4 Data, Measures, andMethods
4.1 Data
Our analyses use data from the 2009 Programme for International Student Assessment
(PISA) survey, collected by the Organization of Economic Cooperation and Development
(OECD). PISA is a nationally representative triennial survey, begun in 2000, that assesses
the academic performance, problem-solving skills, and digital technology use of 15-year-
old students, regardless of grade level. The 2009 survey is uniquely suited to examining the
second digital divide because it includes a variety of questions related to students’ behav-
iors and attitudes regarding digital use. The timing of the 2009 survey also coincides with a
period of increased use of the Internet worldwide. Further, the 2009 student questionnaire
contains a unique set of questions related to Internet literacy, referred to as online reading
(OECD 2012), that are not available from other cycles of the student questionnaire. While
the original sample contains 73 countries, we restrict our analyses to 55 countries because
of missing data on country-level variables. This sample size is large enough to generate
reliable country-level estimates (multilevel models with fewer than 35 countries can yield
underestimated country-level variations) (Bryan and Jenkins 2016).
Using the International Telecommunication Union (ITU 2011, p. 27) categorization,
our paper includes 8 low-income countries, 17 medium-income countries, and 30 upper-
and high-income countries.2 In our analyses, we adopt this income classification, and to
facilitate ease of discussion, we identify the categories as: low-income countries, middle-
income countries, and high-income countries. While our analyses include countries from a
wide range of economic backgrounds, it should be noted that a large proportion of coun-
tries that participated in PISA are from the developed world—a problem in most interna-
tional data sets.
2 The gross national income per capita for low-income countries is below $4570 in U.S. dollars; medium-
income countries are between $4571 and $14,090; and upper- and high-income countries are above
$14,091.
141
National Income, Political Freedom, andInvestments inR&D…
1 3
To preserve cases, we use multiple imputations (m = 10) for missing values in the indi-
vidual-level control variables (Royston et al. 2009). The original sample size of students
across 55 countries is 402,671. Dropping missing cases in the dependent variables and
the key independent variable—family SES—leads to final sample sizes of 391,261 and
398,681 cases for the two dependent variables discussed below. In the “Appendix”, Table5
reports the descriptive statistics for key individual-level variables in each of the 55 coun-
tries,3 and Table6 reports the values of country-level variables for the 55 countries.4
4.2 Dependent Variables
In this research we focus on the association between country-level variables and the sec-
ond digital divide, operationalized as the extent to which family SES affects two proxy
measures for digital use—students’ use of computers for educational purposes at home and
their Internet literacy. Beginning with computer use for educational purposes at home, two
measures are available in PISA. The first is a composite IRT score of digital use for school-
work at home, available for 37 countries. The second is a dichotomous variable of whether
students use educational software at home, available for 55 countries. Both variables are
reasonable proxies for digital use, but the second measure is more narrowly defined. Sup-
plementary analyses using the two measures as dependent variables show similar effect
patterns (see Fig.3 in the “Appendix”). We opt to use the second measure because it allows
us to examine a substantially larger and more diverse sample of countries (55 vs. 37).
Internet literacy, a second proxy for digital use, is measured by a composite scale of
five online reading activities (α = .79): reading online news, using an online dictionary or
encyclopedia, searching online information to learn about a particular topic, taking part in
online group discussions or forums, and searching for practical information online (e.g.,
schedules, events, tips, or recipes).5 We focus on these five items because each has been
found to be positively related to students’ offline reading proficiency (OECD 2011). The
combination of these items indicates student familiarity with reading text on the screen,
sharing information and exchanging ideas, and interacting with others in a digital context.
The variable is standardized with a mean of 0 and a standard deviation of 1.
4.3 Individual‑Level Variables
Our key individual-level independent variable, family SES, is based on the PISA-created
Index of Economic, Social, and, Cultural Status (OECD 2012), which is the most common
measure of SES in studies using PISA data. The variable is a combination of three compo-
nents: (1) parental occupation status, from the international socio-economic index of occu-
pational status (ISEI) (Ganzeboom etal. 1992), (2) parental education in years, and (3) an
index of household possessions, such as a room for the child, owning classical literature, a
3 To account for the possibility that countries with large sample sizes may disproportionately affect param-
eter estimates, we run supplementary analyses with a variable measuring country sample size and find our
results to be unchanged.
4 Following previous studies (Chiu 2010; Decancq and Schokkaert 2016), we present the unlogged values
in the appendix to give the reader a better sense of the actual range of values for the country-level variables.
5 Seven online activities are listed in the questionnaire. We excluded two of these activities—reading
emails and chatting online—as they are less relevant to students’ online literacy and reading proficiency
(OECD 2011).
142
J.K.-H.Ma et al.
1 3
desk for the child to study at home, and the number of books at home. To ease interpreta-
tion of the results, the variable is standardized to have a mean of 0 and a standard deviation
of 1.6
In addition, we include four individual-level control variables that were collected
through the PISA survey. Gender controls for the potential digital gap between male and
female students (male = 1). To control for the effect of immigration status, we include two
dummy variables—first-generation immigrant and second-generation immigrant—with
non-immigrant student as the reference category. To control for differences in language
used by immigrant students, we include a dummy variable—foreign language use at
home—with primary language is the same for home and school as the reference category.
We include this control because students who are not native speakers of the language used
at school may be academically disadvantaged compared to native-speaking students. To
control for family structure, we include two dummy variables—single-parent family and
other family—with two-parent family as the reference category. We assume that omitting
these four control variables induces common-cause confounding bias because they may
affect both our key independent variable (family SES) and our outcome variables (Elwert
and Winship 2014).
4.4 Country‑Level Variables
To examine cross-national differences in the second digital divide, we compile a set of
country-level factors from a variety of publicly available sources. To measure a country’s
national income, we use Gross Domestic Product (GDP) per capita, in thousands of 2009
purchasing power parity dollars, obtained from the World Bank’s World Development
Indicators (WDI) (2015a). We use the composite polity score to measure the level of politi-
cal freedom. This is a combined democracy-autocracy index developed by Marshall etal.
(2010). The scale ranges from − 10 (strongly autocratic) to 10 (strongly democratic). To
measure a country’s investment in R&D and secondary education, we include R&D as a
percentage of GDP from the WDI (2015a) and secondary educational expenditures as a
percentage of GDP from the World Bank’s Education Statistics (2015b). All of the coun-
try-level data are from 2009—the year that the individual-level PISA data were collected.7
Natural log values are used for all country-level variables to account for the skewness of
the distribution (Ruiter and van Tubergen 2009) and to address potential curvilinear rela-
tionships (Heisig 2011).8 Table1 presents descriptive statistics and coding for all variables
used in the analyses.
4.5 Analytical Strategy andStatistical Methods
We use multilevel models to analyze the effects of country-level factors on the two depend-
ent variables and to account for the interdependent variations caused by the clustering of
7 For countries that have missing data on country-level variables in 2009, we utilize data from the closest
adjacent year in which data are available (see Table6 in the Appendix).
8 Because the composite polity score ranges from − 10, to 10, we take a linear transition by adding 11
before logging to ensure that all values are positive.
6 Some studies use the number of books at home as a proxy for family SES or social class (Carnoy and
Rothstein 2013). We consider this alternative in supplementary analyses and find the results to be substan-
tively the same.
143
National Income, Political Freedom, andInvestments inR&D…
1 3
Table 1 Descriptive statistics and variable descriptions in the analysis
Variable Mean/proportion SD Description/coding
Individual-level variables
Use of educational software at home 0.49 1 = yes, 0 = no.
Internet literacy 0.00 1.00 Standardized variable based on five online reading activities (Cronbach’s α = .79): reading
online news, using an online dictionary or encyclopedia (e.g., wikipedia), searching online
information to learn about a particular topic, taking part in online group discussions or
forums, and searching for practical information online (e.g., schedules, events, tips, or reci-
pes). Response categories from lower to higher values are: “I don’t know what it is”, “never or
almost never”, “several times a month”, “several times a week”, and “several times a day”.
Family SES 0.00 1.00 Standardized and PISA-created index of economic, social, and cultural status (OECD 2012),
including: parental occupation status expressed as the index of ISEI, parental education in
years, and an index of household possessions (e.g., a room for the child, possessions of classi-
cal literature, a desk for the child to study at home, and the number of books at home).
Male 0.49 1 = male, 0 = female.
First-generation immigrant 0.04 1 = yes, 0 = no. Reference group = non-immigrant student.
Second-generation immigrant 0.05 1 = yes, 0 = no. Reference group = non-immigrant student.
Foreign language use at home 0.11 1 = yes, 0 = no.
Single-parent family 0.17 1 = yes, 0 = no. Reference group = two-parent family.
Other family 0.04 1 = yes, 0 = no. Reference group = two-parent family.
Country-level variables (all natural log transformed)
GDP per capita 3.08 0.68 Gross domestic product per capita in thousands of 2009 purchasing power parity (PPP) dollars.
The unlogged value ranges from 3.54 to 76.85.
Composite polity score 2.86 0.42 Composite variable based on two variables: democracy and autocracy. The scale ranges from
− 10 (strongly autocratic) to 10 (strongly democratic). Before it is natural log transformed, all
values are transferred into positive integers by adding 11. The unlogged value ranges from − 7
to 10.
R&D as % of GDP − 0.14 1.08 Research and Development including both public and private expenditures that cover basic
research, applied research, and experimental development. The unlogged value ranges from
.05 to 4.15.
144
J.K.-H.Ma et al.
1 3
Table 1 (continued)
Variable Mean/proportion SD Description/coding
Secondary educational expenditures as % of GDP 0.56 0.40 Total government expenditures on secondary education from the local, regional, and central
government and transfers from international sources. The unlogged value ranges from .64 to
3.63.
Data Source All individual-level variables are from the Programme for International Student Assessment (PISA) 2009; GDP per capita and R&D are compiled from the
World Bank’s World Development Indicators (2015a); composite polity score is from Marshall etal.’s (2010) Polity IV Project; secondary educational expenditures are from
the World Bank’s Education Statistics (2015b)
To preserve cases, multiple imputations (m = 10) for missing cases are used for individual-level control variables
145
National Income, Political Freedom, andInvestments inR&D…
1 3
students within countries (Rabe-Hesketh and Skrondal 2008). The multilevel analysis con-
sists of an individual- and a country-level model. At the individual-level, the general form
of the models for a student i in country j can be written as,
The left-hand side link functions
𝜂ij
are treated differently for binary and continuous out-
come variables. For the binary dependent variable—use of educational software at home—
𝜂ij
can be specified as:
where
∅i
is equal to P(y = 1|X), making the model a multilevel logistic model. For our con-
tinuous dependent variable—Internet literacy—
𝜂ij
is equal to y.
𝛽
0j is the individual-level
intercept, adjusted for family SES and other individual-level control variables.
𝛽1j
is the
coefficient of family SES.
rij
is the unexplained variance for individual i in country j. At the
country-level, we assume:
where the intercept and the coefficient to family SES slope are allowed to randomly vary
across nations.9
Z
1
j
to
Zkj
indicate a set of country-level variables. All continuous country-
level variables are centered at the grand mean, so that
𝛾00
represents the grand mean of the
intercept and
𝛾10
indicates the grand mean of the family SES slope for countries whose
country-level variables are set at the average values. The main focus of this paper is to
examine the association between national contextual factors and the digital divide, meas-
ured as the slope of family SES regressed on the two outcome variables (Eq.3).
Our analyses proceed in three stages. In the first stage, we use logit and linear regres-
sions, respectively, for educational software use at home and Internet literacy, and estimate
the models separately in each of the 55 countries. Based on these models, we use graphs to
visualize how the effect of family SES on the two digital use measures varies across coun-
tries of different economic standings. Next, we use multilevel models to formally examine
the country-level variation in the second digital divide. To test our hypotheses that national
income, political freedom, R&D investment, and secondary educational expenditures
reduce the second digital divide, we estimate the effects of these national indicators on the
two dependent variables and the slopes of family SES. Finally, we examine whether the
(1)
𝜂
ij =𝛽0j+𝛽1j(Family SES)ij +
k
∑
2
𝛽kjXkij +r
ij
𝜂
ij =ln
(�
i
1−�
i),
(2)
𝛽
0j=𝛾00 +
k
∑
1
𝛾0kZkj +𝜇0
j
(3)
𝛽
1j=𝛾10 +
k
∑
1
𝛾1kZkj +𝜇1
j
9 To avoid over-parameterization, we do not consider the random slopes of other individual-level variables.
In models not shown, we find the inclusion of additional random slopes does not influence the main results
reported here. Additionally, we find family SES to be much more important than other variables in captur-
ing cross-cluster heterogeneity. These models are available upon request.
146
J.K.-H.Ma et al.
1 3
effects of national investments in R&D and secondary education on the family SES slopes
differ across low-, middle-, and high-income countries. Based on the estimated models, we
calculate the predicted SES slopes for the 55 countries and present the results in graphs.
5 Results
Figure1 illustrates the variations in the second digital divide (more specifically, the rela-
tionship between family SES and our two proxy measures for digital use) across three
country-level income groups using the results of separate regression models for each
country. Overall, the effect of family SES on the two outcomes differs substantially across
national income levels. On average, the slopes of family SES among low-income coun-
tries are steeper than the slopes in most middle- and high-income countries (for example,
compare the SES slopes between Georgia, Lithuania, and Spain when predicting Internet
literacy). This suggests that the second digital divide is more pronounced in poor countries.
However, we also see that the slopes of family SES vary among countries with similar
wealth, particularly among high-income countries. This suggests that national income can
only partially explain cross-national variation in the level of the second digital divide.
Using multilevel modeling, we more formally examine country-level variation in the
second digital divide in Table2. We first note that the intra-class correlation coefficients
(ICC) for empty models (models that only include the intercept) are .127 when predict-
ing educational software use at home, and .114 when predicting Internet literacy. In other
words, about 11–13% of the variation in the intercept occurs at the country level. While
informative, these numbers can only represent the cross-national variation in the intercept.
In what follows, we examine how family SES slope—the main focus of our study—varies
across countries.
Model 1 shows the effect of family SES on the use of educational software at home
in all nations. Among students with average family SES, the predicted probability of
using educational software at home is 50% (= 1/(1 + e{−(−.015 + 0*.823)})) and for a
one-standard-deviation increase in SES, the probability increases to 69% (= 1/(1 + e{−
(−.015 + 1*.823)})). Model 1 also reveals that the country-level variance in family SES
is .056 (p < .01), which suggests a 95% confidence interval ranging from .358 to 1.288
(= .823 ± 1.96*
√.056
) for the SES coefficients. In substantive terms, this means that,
excluding the extreme 5% of the two sides, the predicted probabilities range from 58%
(= 1/(1 + e{−(−.015 + .358)})) to 78% (= 1/(1 + e{−(−.015 + 1.288)})) for students whose
family SES is one standard deviation above the mean. This level of variation justifies an
examination of the influence of SES on digital use across the 55 countries in this analysis.
Model 2 introduces the individual-level control variables. The estimated effect of family
SES decreases only slightly and remains statistically significant (
𝛽
= .816, p < .01). A one-
standard-deviation increase in family SES increases the probability of using educational
software at home by 19% (= 1/(1 + e{−(.097 + .816)})−1/(1 + e{−(.097)})). Results from
the variance components show that the 95% confidence interval for the SES slope ranges
from .364 to 1.268, indicating that a one-standard-deviation increase of SES will increase
the probability of student use of educational software at home by between 9% and 27%
across the 55 countries.
Model 3 presents the effect of family SES on Internet literacy. For each standard
deviation increase in family SES, Internet literacy increases by .338 standard devia-
tions (p < .01). After including other individual-level characteristics in Model 4, the
147
National Income, Political Freedom, andInvestments inR&D…
1 3
Fig. 1 Regression lines of educational software use at home and Internet literacy. Note The left panel con-
tains predicted plots based on logit models. The right panel contains predicted plots based on OLS models.
Each model includes individual-level control variables (gender, immigration status, foreign language use at
home, and family structure)
148
J.K.-H.Ma et al.
1 3
coefficient for family SES increases slightly to .342. When taking variance components
into account, we find that the effect of family SES on Internet literacy ranges between
.093 and .591 standard deviations among 95% of the countries in our analysis. Taken
together, we conclude that family SES significantly affects both educational software
use at home and Internet literacy, but we note that the size of the effect varies substan-
tially across countries (Model 2:
𝜇
1
j
= .053, p < .01, Model 4:
𝜇
1
j
= .016, p < .01). For
instance, the effect of family SES in Jordan (SES slope = .542) is approximately four
times greater than in Portugal (SES slope = .137). This degree of cross-country variation
warrants further investigation and raises the question: which country-level indicators are
associated with the digital divide? This is a question with obvious policy implications,
Table 2 Multilevel analyses for the second digital divide with individual-level variables
Note Number of countries = 55. Robust standard errors are in parentheses. All coefficients are adjusted by
multiple imputations for missing cases in the control variables (m = 10). Log-likelihood is from imputed
dataset m = 1
a For an intercept-only model: between-country intercept variance is .479. The intraclass correlation (ICC)
is .127
b For an intercept-only model: between-country intercept variance is .117. Within-country variance is .913.
ICC is .114
**p < .01, *p < .05, + p < .1 (2-tailed)
Use of educational software at
homea
Internet literacyb
Model 1 Model 2 Model 3 Model 4
Intercept − .015 (.082) .097 (.086) .034 (.036) − .012 (.038)
Family SES .823 (.032)** .816 (.032)** .338 (.018)** .342 (.017)**
Male − .139 (.024)** .070 (.010)**
Reference group: female
Immigration status
Reference group: non-immigrant
First-generation immigrant .075 (.040)+ .163 (.033)**
Second-generation immigrant .217 (.053)** .205 (.026)**
Foreign language use at home − .073 (.047) − .030 (.038)
Reference group: primary language is
the same for home and school
Family structure
Reference group: two-parent family
Single-parent family − .224 (.017)** − .003 (.006)
Other family − .229 (.038)** − .067 (.018)**
Variance components
Between-country intercept variance .364** .360** .072** .071**
Between-country family SES variance .056** .053** .017** .016**
Within-country variance .813 .810
Log-likelihood − 5,53,994 − 5,54,264 − 5,24,719 − 5,23,907
NIndividual-level 391,261 391,261 398,681 398,681
149
National Income, Political Freedom, andInvestments inR&D…
1 3
and we will attempt to address it in the next two sections, but not before first reviewing
the effects of the remaining individual-level controls included in these models.
Most of the individual-level controls in Models 2 and 4 perform as expected. When con-
trolling for SES and other socio-demographic characteristics, students who are second-gen-
eration immigrants have the greatest likelihood of using educational software at home and
have the highest level of Internet literacy compared to first-generation students (p < .01,
results not shown) and non-immigrants (p < .01). Non-immigrant students have the lowest
level of digital use for both outcome variables. Also, speaking a foreign language at home
does not significantly affect educational software use at home or Internet literacy. These
findings correspond to recent literature that suggests a “reverse digital divide” between
immigrants and nonimmigrants (Milioni etal. 2014), but future researchers should further
investigate whether this reverse digital divide is observed across different regions of the
world.
Compared to two-parent families, students living in single-parent families are less likely
to use educational software at home, but have similar levels of Internet literacy. Finally,
we note that male students have significantly higher Internet literacy than females, which
is consistent with previous studies that have found a tendency for girls to report lower self-
assessment of online skills than boys (Hargittai and Shafer 2006). Despite this gender gap
in Internet literacy that advantages boys, male students are less likely to report using edu-
cational software at home than females (Model 2). This result corresponds with findings
from recent studies which find that despite their high rates of computer use, male students
are more likely than female students to use computers for non-educational activities, such
as gaming (Imhof etal. 2007).
5.1 Sources ofCross‑National Variation intheSecond Digital Divide
To evaluate cross-national variation in the second digital divide, we estimate multilevel
models assessing the effects of country-level variables on the two outcomes as well as the
slope of family SES. In Table3, we first include our measures of national income (GDP
per capita) and political freedom (composite polity score) to examine whether the associa-
tion between economic factors and the second digital divide can be explained by political
factors, or vice versa (Model 1 for educational software use at home; Model 4 for Internet
literacy). We then test whether R&D and secondary educational expenditures as a percent
of GDP affect the second digital divide in use of educational software at home (Models 2
and 3) and Internet literacy (Models 5 and 6).10 We continue to control for national income
and political freedom in these models in order to test whether the effects of R&D and edu-
cational expenditures persist net of economic and political forces. All analyses include the
same individual-level control variables shown in Table2. The top half of Table3 shows the
effects of country-level measures on the intercept. The bottom half of the table examines
the effects of country-level variables on the slope of family SES (the level of the second
digital divide). The differential effects of family SES by R&D and secondary educational
expenditures across country-level income groups are further examined in Table4.
Beginning with educational software use at home, we see in Model 1 that the average
effect of family SES is .815. This indicates that a one-standard-deviation increase in family
SES increases the probability of using educational software by 19%. The coefficients for
10 In supplementary analyses including both R&D and educational expenditures, the significant effect of
R&D disappears because of the high correlation between the two variables (r = .48), but the general pat-
terns remain the same.
150
J.K.-H.Ma et al.
1 3
Table 3 Multilevel analyses for the second digital divide with country-level variables
Note Number of countries = 55. Robust standard errors are in parentheses. All coefficients are adjusted by multiple imputations for missing cases in the control variables
(m = 10). Log-likelihood is from imputed dataset m = 1. All models include individual-level control variables (gender, immigration status, foreign language use at home, and
family structure). Family SES is group mean centered. All country-level variables are natural log transformed and grand mean centered
**p < .01, *p < .05, + p < .1 (2-tailed)
Use of educational software at home Internet literacy
Model 1 Model 2 Model 3 Model 4 Model 5 Model 6
Effects on the intercept
Intercept .095 (.087) .095 (.087) .095 (.083) − .019 (.043) − .019 (.043) − .019 (.042)
GDP per capita .703 (.113)** .684 (.103)** .615 (.104)** .183 (.068)** .156 (.078)* .155 (.067)*
Composite polity score .117 (.115) .103 (.114) − .123 (.191) .039 (.091) .020 (.090) − .036 (.125)
R&D as % of GDP .025 (.095) .035 (.035)
Secondary educational expenditures as % of GDP .612 (.250)* .193 (.127)
Effects on the family SES slope
Intercept .815 (.027)** .815 (.025)** .815 (.025)** .341 (.012)** .341 (.011)** .341 (.012)**
GDP per capita − .193 (.041)** − .139 (.041)** − .164 (.036)** − .138 (.022)** − .112 (.023)** − .135 (.024)**
Composite polity score − .022 (.035) .017 (.045) .058 (.050) − .033 (.035) − .015 (.026) − .026 (.035)
R&D as % of GDP − .071 (.029)* − .033 (.009)**
Secondary educational expenditures as % of GDP − .206 (.058)** − .017 (.033)
Variance components
Between-country intercept variance .395** .394** .349** .097** .096** .093**
Between-country family SES variance .036** .032** .031** .007** .006** .007**
Within-country variance .810 .810 .810
Log-likelihood − 5,54,259 − 5,54,265 − 5,54,259 − 5,23,894 − 5,23,890 − 5,23,892
NIndividual-level 391,261 391,261 391,261 398,681 398,681 398,681
151
National Income, Political Freedom, andInvestments inR&D…
1 3
Table 4 Multilevel analyses for the second digital divide: differential effects by country income group
Use of educational software at home Internet literacy
Model 1 Model 2 Model 3 Model 4
Effects on the Intercept
Intercept .495 (.086)** .368 (.129)** .048 (.046) .059 (.045)
Low-income countrya− 1.004 (.502)* − 1.120 (.446)* − .522 (.268)+ − .484 (.152)**
Middle-income countrya− .358 (.249) − .340 (.387) .001 (.149) − .228 (.117)+
Composite polity score .066 (.167) − .394 (.232)+ − .048 (.107) − .107 (.104)
R&D as % of GDP − .108 (.083) .033 (.036)
R&D as % of GDP × Low-income country .474 (.339) − .083 (.191)
R&D as % of GDP × Middle-income country .339 (.270) .031 (.131)
Secondary educational expenditures as % of GDP .742 (.529) .069 (.141)
Secondary educational expenditures as % of GDP × Low-income country − .346 (.626) .060 (.273)
Secondary educational expenditures as % of GDP × Middle-income country .077 (.589) .404 (.242) +
Effects on the family SES slope
Intercept .720 (.048)** .729 (.042)** .282 (.011)** .275 (.013)**
Low-income countrya.106 (.051)* .302 (.109)** .272 (.082)** .123 (.043)**
Middle-income countrya.093 (.056) + .124 (.102) .117 (.025)** .077 (.032)*
Composite polity score .001 (.057) .118 (.063) + .029 (.023) .068 (.022)**
R&D as % of GDP − .026 (.050) − .041 (.008)**
R&D as % of GDP × Low-income country − .198 (.062)** .065 (.054)
R&D as % of GDP × Middle-income country − .089 (.073) .031 (.031)
Secondary educational expenditures as % of GDP − .228 (.116)* − .145 (.036)**
Secondary educational expenditures as % of GDP × Low-income country .073 (.135) .287 (.054)**
Secondary educational expenditures as % of GDP × Middle-income country .003 (.120) .069 (.048)
Variance components
Between-country intercept variance .368** .339** .086** .079**
Between-country family SES variance .029** .030** .006** .005**
152
J.K.-H.Ma et al.
1 3
Table 4 (continued)
Use of educational software at home Internet literacy
Model 1 Model 2 Model 3 Model 4
Within-country variance .810 .810
Log-likelihood − 5,54,287 − 5,54,265 − 5,23,885 − 5,23,879
NIndividual-level 391,261 391,261 398,681 398,681
Note Number of countries = 55. Robust standard errors are in parentheses. All coefficients are adjusted by multiple imputations for missing cases in the control variables
(m = 10). Log-likelihood is from imputed dataset m = 1. All models include individual-level control variables (gender, immigration status, foreign language use at home, and
family structure). Family SES is group mean centered. R&D as % of GDP and Secondary educational expenditures as % of GDP are natural log transformed and grand mean
centered
a High-income country is the reference category
**p < .01, *p < .05, +p < .1 (2-tailed)
153
National Income, Political Freedom, andInvestments inR&D…
1 3
GDP per capita in the intercept and the family SES slope equations are .703 and − .193,
respectively. This suggests that national income is associated with an increase in the prob-
ability of using educational software and, at the same time, a reduction in the socioeco-
nomic disparity in digital use. For example, a one-standard-deviation increase in family
SES increases the probability of using educational software by 13% for low-income coun-
tries (e.g., India: GDP per capita is $3960 in U.S. dollars), compared to only 2% for high-
income countries (e.g., Netherlands: GDP per capita is $44,400 in U.S. dollars). The pol-
ity score has no significant effect (p > .1). Together, these findings confirm that national
income outweighs political freedom when explaining the digital divide (Robison and Cren-
shaw 2010).
In Model 2 we find that increasing national investment in R&D does not affect educa-
tional software use at home, but reduces the socioeconomic gap in its use. The magnitude
of the effect is moderate, with a one-log-unit increase in R&D leading to a 19% decrease
in the standardized effect of family SES, which is approximately the difference between
Serbia (R&D = .87% of GDP) and Australia (R&D = 2.4% of GDP). Model 3 shows that
students living in countries with higher secondary education expenditures are more likely
to use educational software at home (b = .612, p < .05). More importantly, the negative
association between educational expenditures and the digital divide is both statistically sig-
nificant and substantial in size, with a one-unit-increase in expenditures leading to a 25%
decrease in the standardized effect of family SES. To place this in the context of our data,
a one-standard-deviation increase in family SES increases the probability of using educa-
tional software by 21% for countries with lower educational expenditures (e.g., Thailand,
where educational expenditures make up 0.64% of GDP), compared to 15% for countries
with higher educational expenditures (e.g., Brazil, 2.51% of GDP).
Models 4, 5, and 6 examine the effects of country-level variables on Internet literacy.
In Model 4, we see that a one-standard-deviation increase in family SES increases Inter-
net literacy by .341 standard deviations, holding individual-level variables constant. GDP
per capita increases students’ Internet literacy (b = .183, p < .05) and reduces the Internet
literacy gap (b = − .138, p < .01). The polity score is not associated with Internet literacy
or the slope of family SES. Model 5 shows that R&D does not have a significant effect on
Internet literacy, but significantly reduces the digital divide in Internet literacy (b = − .033,
p < .01). In Model 6, we see that educational expenditures as a percent of GDP do not
have a significant relationship with Internet literacy or the second digital divide in Internet
literacy.
Overall, the results in Table3 suggest that national income is negatively associated with
the second digital divide for teenage students. Net of economic and political factors, invest-
ments in R&D and secondary education are associated with reductions in the second digi-
tal divide, though the effects of educational expenditures are limited to the use of educa-
tional software at home. However, these general patterns may vary across different levels
of national income—a possibility we examine next.
5.2 Dierential Eects Between Low‑, Middle‑, andHigh‑Income Countries
Table4 reports the differential effects of R&D investment and secondary education expen-
ditures on the second digital divide between low-, middle-, and high-income countries.
Based on Hypotheses 5a and 5b discussed earlier, we focus on the interaction effects
between the key independent variables and the level of national income on the slope
of family SES. We report two notable interaction effects. First, Model 1 shows that the
154
J.K.-H.Ma et al.
1 3
magnitude of the effect of family SES on educational software use at home is greater for
low-income countries (b = .720+.106) than high-income countries (b = .720). Moreover,
R&D as a percent of GDP is negatively associated with the family SES slope, but the rela-
tionship is much stronger for low-income countries (b = − .026–.198) than high-income
countries (b = − .026). Second, Model 4 suggests that the association between second-
ary educational expenditures and the family SES slope when predicting Internet literacy
is negative for high-income countries (b = − .145, p < .01) and moderately negative for
middle-income countries (b = − .145 +.069, p < .10), but positive for low-income countries
(b = − .145 +.287, p < .01). In other words, the higher the national income, the greater the
Fig. 2 Predicted family SES slopes by country income group. Note Predicted family SES slopes are calcu-
lated from Table4 (Model 1 in the top left corner; Model 2 in the bottom left; Model 3 in the top right; and
Model 4 in the bottom right). The plotted lines represent the association between the variable on the x-axis
and the slope of family SES for low-income (dashed line), middle-income (dotted line), and high-income
countries (solid line). The symbols attached to country acronyms represent the predicted family SES slopes
adjusted for between-country variance. Both R&D and secondary educational expenditures, measured as %
of GDP, are natural log transformed and centered at the grand mean. 0 is the mean of 55 countries. All coef-
ficients are adjusted by multiple imputations for missing cases in the control variables (m = 10)
155
National Income, Political Freedom, andInvestments inR&D…
1 3
negative association between educational expenditures and the socioeconomic divide in
Internet literacy.
To further examine the patterns of these differential effects, we plot the predicted slopes
of family SES by national income in Fig.2. Each graph presents the expected level of
the digital divide pertinent to the relationship between family SES and the outcome vari-
ables of interest (y-axis), conditional on different national income levels with the change of
R&D (x-axis on the top charts) or educational expenditures (x-axis on the bottom charts).
Beginning with educational software use at home (the left panel), we find a strong negative
relationship between R&D and the SES slope for low- and middle-income countries, but
not for high-income countries. This finding suggests that investment in R&D can play an
important role in reducing the second digital divide in lower-income countries, but it offers
little advantage in high-income countries. We should note that low-income countries’ R&D
investments as a percent of GDP are smaller than most high-income countries, which
leaves a lot of room for R&D growth and hence reduction of the second digital divide
for these countries. Conversely, the relatively small effect found in high-income countries
suggests there may be a ceiling above which further R&D spending is no longer helpful
in reducing digital inequality. We also find a negative association between secondary edu-
cational expenditures and the digital divide—a relationship that persists across national
income levels. This finding suggests that educational expenditures play a role in reducing
the digital divide in software use at home in all countries, regardless of national wealth.
Moving to Internet literacy, we also find notably different effects for SES across coun-
tries with different levels of national income. Overall, the negative association between
R&D and educational expenditures and levels of the digital divide are stronger in high-
income countries. First, increased R&D spending is associated with a decline in the effect
of SES among high-income countries, but not in low- or middle-income countries.11 This
may be due in part to a lack of Internet access in lower-income countries, which can restrict
the role of R&D in bridging the second digital divide in Internet literacy. Second, the neg-
ative relationship between educational expenditures and the effect of SES exists only in
high-income countries, while the effect actually becomes positive when looking at low-
income countries. In other words, increased educational expenditures are associated with a
widening digital divide in Internet literacy among poor nations. This counterintuitive find-
ing raises the question: why don’t educational expenditures lead to greater equality in digi-
tal literacy for students in low-income countries?
We consider several possible answers to this question through a series of supplementary
analyses.12 First, because of the small number of countries, it is possible that the patterns
for low-income countries observed in Fig.2 are sensitive to the categorization scheme. To
test the robustness of the findings, we repeat the same analyses in Table4 and Fig.2 using
an alternative classification of country-level income groups derived from the World Bank
(2015c).13 The conclusion remains consistent.
11 Trinidad & Tobago and Macao are potential outliers in Fig.2 with regard to the effect of R&D for high-
income nations. Supplementary analyses excluding these two countries show patterns consistent with those
reported here.
12 Results from supplementary analyses are not shown here, but are available from the authors upon
request.
13 Based on the World Bank, there are 6 lower-middle-income countries, 15 upper-middle-income coun-
tries, and 34 high-income countries.
156
J.K.-H.Ma et al.
1 3
Second, it is possible that the high levels of income inequality in less-developed nations
may contribute to the widening digital divide. To account for this possibility, we run sup-
plemental analyses including the Gini index as a covariate. We find increased income ine-
quality widens the digital divide (p < .01), but the size of this effect is reduced and becomes
statistically non-significant once controlling for national income. More importantly, the
levels of income inequality among low-income countries (e.g., Moldova and Tunisia,
whose Gini indices are .39 and .41, respectively) are lower on average than most middle-
income countries (e.g., Colombia and Brazil, whose Gini index is .56), which suggests that
income inequality itself does not explain the widening digital divide in poor countries.
Finally, we consider the possibility that rates of secondary educational enrollment are
associated with the level of digital inequality in Internet literacy.14 We find that higher
enrollment rates are associated with greater inequality among less-developed countries, but
this association does not exist in developed countries. Together, the results of these analy-
ses suggest the benefits of increased educational spending are disproportionately enjoyed
by socioeconomically advantaged students in low-income countries. Unlike the efforts of
governments in affluent countries to promote educational equity, educational planning in
poor countries may prioritize the needs of students who are already socioeconomically
privileged while leaving a large number of less-affluent students behind.
To summarize, our analyses in Table3 show that national income, R&D investments,
and educational expenditures have significant effects on reducing the second digital divide
among teenage students, providing support for Hypotheses 1, 3, and 4 as put forth earlier.
Table4 and Fig. 2 further indicate significant differences in the effect of national invest-
ments in R&D and educational expenditures in bridging the second digital divide across
countries with various income levels. Consistent with previous literature that questions the
role of educational expenditures in reducing educational inequalities (Hannum and Buch-
mann 2005; Juárez and Gayet 2014), our findings indicate that, among less-developed
countries, the benefits of increased educational spending may be limited to only the most
affluent students. In contrast, increased educational spending in developed countries is
more beneficial to less-affluent students (Hypothesis 5a).
6 Discussion andConclusion
The use of digital technology in education has continued to grow in the past decade, mak-
ing digital literacy an increasingly important component of success for students. However,
despite its growing importance for education, a digital divide in technology use persists
worldwide. Based on PISA reports, the disparity in digital access—referred to as the first
digital divide—has narrowed in most countries, while the disparity in digital use between
students from varying socioeconomic backgrounds remains substantial (OECD 2011,
2015). Findings from the PISA 2009 digital reading assessments for students across 19
countries suggest that SES explains about 14% of the variation in students’ digital read-
ing performance, which is equivalent to a gap of over 2 years of schooling (OECD 2011,
p. 124). According to PISA 2012 data, low-SES students have weak computer navigation
skills and online literacy compared to high-SES students and are less likely to start using
14 The Gini index data are obtained from the World Income Inequality Database (UNU-WIDER 2008).
Secondary education enrollment rates are derived from the World Bank (2015a).
157
National Income, Political Freedom, andInvestments inR&D…
1 3
computers for learning in early childhood because their families “may not be aware of how
technology can help to raise one’s social status (OECD 2015, p. 125).” This socioeconomic
disparity in digital use, known as the second digital divide, has been identified in every
country where data have been collected (Notten etal. 2009).
Despite the wealth of research on the digital divide, national-level factors that contribute
to the second digital divide among students have received limited attention. This is surpris-
ing, since scholars have long recognized the need to investigate the role of public policies
in the integration of e-learning into schools and education (DiMaggio etal. 2004; Erichsen
and Salajan 2014; Natriello 2001). Motivated by this gap in the research as well as findings
from previous studies (Norris 2001), we investigate how national income, political free-
dom, and national investments in R&D and secondary education are associated with the
second digital divide among 15-year-old students in 55 countries.
Our analyses reveal several key findings. First, we find national income to be a power-
ful predictor of the second digital divide among teenage students. Additionally, national
investments in R&D and secondary education are negatively associated with the second
digital divide. The size of the relationships for these measures are modest, but they remain
statistically significant even after controlling for national income and individual-level
background characteristics. Given these findings, we surmise that targeted investments in
research, innovation, and education aimed at enhancing digital learning opportunities for
all students could potentially reduce digital inequality. Policymakers interested in reducing
digital inequality may want to consider this finding when constructing policies to address
the digital divide.
Equally important, we find that the economic standing of countries shapes how different
national investments may influence digital inequality among youths. For example, R&D
spending reduces the second digital divide in educational software use at home, but only
in less-developed countries, suggesting there may be an opportunity to reduce the digital
divide for these countries since they have the greatest room to expand their investments in
R&D (see Table6 in the “Appendix” for examples). Moreover, increased R&D spending
and educational expenditures are associated with reducing the Internet literacy gap among
high-income countries, but not in low- or middle-income countries. In fact, and perhaps
surprisingly, increased expenditures on education in low-income countries lead to a widen-
ing Internet literacy gap between lower-SES and higher-SES students. This may be attrib-
utable to the complex interaction of several socio-economic factors such as lack of social
mobility, weak labor markets, and widespread poverty in less-affluent countries. We find
some evidence to support this possibility in our data. For example, among the low-income
countries in our sample, Moldova has the highest level of educational expenditures, but its
strikingly poor economic conditions have further exacerbated the hardships of socioeco-
nomically disadvantaged students (Worden 2014).
Despite these significant findings, we acknowledge several shortcomings in this study
and recommend future directions for research. First, since the PISA survey focuses over-
whelmingly on high-income countries—a common problem with international datasets
(Chiu 2010; Park and Kyei 2011)— the number of less-developed countries is limited in
our analysis. Future efforts to collect international comparative data from a more diverse
array of countries will help to reduce this problem. Future scholars should pay special
attention to data from the developing world, as our study indicates the importance of look-
ing at the differences in effects between countries of different income levels. Second, we
note our inability to ascertain the causal relationship between educational expenditures and
158
J.K.-H.Ma et al.
1 3
the level of the second digital divide.15 To establish that connection, future research may
draw from longitudinal data to examine the change in expenditures and the level of digital
inequality across time. Third, our multilevel analyses allow family SES to vary randomly
across country-level clusters, but overlook the random effects of other individual-level con-
trols. Future research should consider using two-step approaches that contain a flexible ran-
dom-effects structure which may yield more precise estimates (Heisig etal. 2017).
Finally, future qualitative and quantitative research should explore specific projects, pol-
icies, or practices that directly or indirectly promote digital literacy—especially among eco-
nomically disadvantaged students. This includes studying the influence of other national-
level indicators from sources outside of the PISA dataset, such as the Human Development
Index by the United Nations Development Programme, social welfare expenditures from
the World Bank’s World Development Indicators, and other educational indices created by
the World Bank’s Education Statistics, to name a few possibilities. Since digital technology
appears likely to be a dominant force in society for the foreseeable future—affecting earn-
ings and other social outcomes—ensuring the next generation is digitally literate should be
a priority for countries seeking to compete in the global economy. So long as a high level
of inequality in digital use persists, social scientists must continue to seek out solutions
by exploring various local and national investments which governments can make to help
reduce the second digital divide.
Acknowledgements The authors would like to thank Brian Powell, Michael Wallace, Jeremy Pais, Mary
Fischer, David Weakliem, Thung-Hong Lin, and several anonymous reviewers for valuable comments on
previous versions of this paper.
Appendix
See Tables5 and 6 and Fig.3.
15 While educational expenditures have a clear and direct relationship to students, countries may distribute
their educational resources in ways that are unrelated to digital technology use. The distribution of educa-
tional investments within a country could also be biased by social status, with newer technologies going
only to schools in the most affluent areas. Bearing these possibilities in mind, we examine whether or not
these investments, when used appropriately, may serve as tools to reduce the second digital divide.
159
National Income, Political Freedom, andInvestments inR&D…
1 3
Table 5 Sample size and descriptive statistics for key individual-level variables in 55 countries
Country Sample size Use of educa-
tional software at
home
Internet literacy Family SES
Mean SD Mean SD Mean SD
Argentina (AR) 4774 .40 .49 − .27 .97 − .19 1.02
Australia (AU) 14,251 .70 .46 − .01 .86 .60 .66
Austria (AT) 6590 .49 .50 .13 .87 .40 .72
Belgium (BE) 8501 .64 .48 − .20 .81 .51 .80
Brazil (BR) 20,127 .24 .43 − .37 1.12 − .70 1.05
Bulgaria (BG) 4507 .62 .48 .45 1.22 .23 .86
Chile (CL) 5669 .37 .48 .00 .97 − .11 1.01
Colombia (CO) 7921 .37 .48 − .07 .96 − .55 1.08
Costa Rica (CR) 4578 .41 .49 − .46 .96 − .48 1.14
Czech (CZ) 6064 .64 .48 .66 .88 .34 .64
Denmark (DK) 5924 .73 .44 .21 .83 .44 .81
Estonia (EE) 4727 .69 .46 .61 .86 .48 .69
Finland (FI) 5810 .35 .48 .02 .81 .68 .68
France (FR) 4298 .39 .49 − .02 .84 .23 .73
Georgia (GE) 4646 .27 .44 − .03 1.14 .10 .85
Germany (DE) 4979 .57 .49 .23 .85 .48 .79
Greece (GR) 4969 .42 .49 .08 1.02 .35 .86
Hong Kong (HK) 4837 .57 .49 .58 .84 − .38 .88
Hungary (HU) 4605 .47 .50 .53 .95 .19 .82
India (IN) 4826 .17 .38 − .90 1.27 − 1.11 .96
Indonesia (ID) 5136 .16 .36 − .77 .97 − 1.00 .95
Ireland (IE) 3937 .57 .49 − .35 .85 .38 .74
Israel (IL) 5761 .54 .50 .26 .96 .32 .77
Italy (IT) 30,905 .54 .50 .11 1.00 .24 .85
Japan (JP) 6088 .16 .36 − .30 .95 .32 .63
Jordan (JO) 6486 .53 .50 − .32 1.14 − .10 .89
Korea (KR) 4989 .60 .49 .21 .85 .21 .71
Latvia (LV) 4502 .70 .46 .53 .89 .29 .75
Lithuania (LT) 4528 .61 .49 .71 .95 .30 .84
Macao (MO) 5952 .62 .49 .13 .85 − .28 .75
Malaysia (MY) 4999 .56 .50 − .40 1.00 − .10 .77
Mauritius (MU) 4654 .59 .49 − .16 1.09 − .25 .87
Mexico (MX) 38,250 .32 .47 − .14 .91 − .67 1.11
Moldova (MD) 5194 .35 .48 .12 1.25 − .16 .85
Netherlands (NL) 4760 .70 .46 − .01 .82 .60 .74
New Zealand (NZ) 4643 .63 .48 − .09 .85 .41 .68
Norway (NO) 4660 .68 .47 .27 .83 .74 .64
Panama (PA) 3969 .31 .46 − .17 1.14 − .33 1.11
Peru (PE) 5985 .31 .46 − .27 1.02 − .80 1.07
Poland (PL) 4917 .74 .44 .71 .94 .14 .79
Portugal (PT) 6298 .65 .48 .24 .87 .07 1.01
Serbia (RS) 5523 .51 .50 − .06 1.08 .39 .83
160
J.K.-H.Ma et al.
1 3
Table 5 (continued)
Country Sample size Use of educa-
tional software at
home
Internet literacy Family SES
Mean SD Mean SD Mean SD
Shanghai (CN) 5115 .40 .49 − .06 .89 − .09 .91
Singapore (SG) 5283 .64 .48 .27 .95 − .04 .70
Slovakia (SK) 4555 .60 .49 .18 .97 .25 .73
Slovenia (SI) 6155 .72 .45 .30 .92 .28 .76
Spain (ES) 25,887 .52 .50 .00 .87 .11 .91
Sweden (SE) 4567 .58 .49 .13 .84 .62 .70
Switzerland (CH) 11,812 .51 .50 .06 .84 .35 .74
Thailand (TH) 6225 .32 .47 − .28 1.04 − .69 1.09
Trinidad & Tobago (TT) 4778 .63 .48 − .24 1.00 − .16 .82
Tunisia (TN) 4955 .32 .47 − .55 1.15 − .74 1.13
United Kingdom (GB) 12,179 .70 .46 .10 .89 .48 .68
United States (US) 5233 .60 .49 .00 .96 .46 .80
Uruguay (UY) 5957 .45 .50 − .01 1.01 − .32 1.07
161
National Income, Political Freedom, andInvestments inR&D…
1 3
Table 6 Country-level variables: 55 countries
GDP per capita Composite
polity score
R&D as %
of GDP
Secondary educational
expenditures as % of
GDP
Low-income countries
Moldova (MD) 3.54 8.00 .53 3.63
India (IN) 3.96 9.00 .82 1.12
Georgia (GE) 5.46 6.00 .18 1.08
Indonesia (ID) 7.82 8.00 .08 .90
Peru (PE) 8.93 9.00 .16 1.05
Tunisia (TN) 9.96 − 4.00 .71 2.97
Jordan (JO) 10.88 − 3.00 .43 1.77
Thailand (TH) 12.26 4.00 .25 .64
Mean 7.85 4.63 .39 1.65
Middle-income countries
Shanghai (CN) 10.13 − 7.00 1.68 .71
Colombia (CO) 10.26 7.00 .21 1.66
Serbia (RS) 11.81 8.00 .87 1.12
Costa Rica (CR) 11.82 10.00 .54 1.36
Brazil (BR) 13.09 8.00 1.12 2.51
Mexico (MX) 13.91 8.00 .43 1.58
Panama (PA) 14.10 9.00 .14 1.10
Mauritius (MU) 14.54 10.00 .37 1.51
Argentina (AR) 14.60 8.00 .48 1.99
Bulgaria (BG) 14.88 9.00 .51 1.80
Uruguay (UY) 15.39 10.00 .42 1.04
Chile (CL) 16.23 10.00 .35 1.49
Latvia (LV) 17.04 8.00 .45 2.27
Lithuania (LT) 18.28 10.00 .83 2.95
Poland (PL) 19.15 10.00 .67 1.85
Malaysia (MY) 19.33 6.00 1.01 1.96
Hungary (HU) 20.87 10.00 1.14 2.04
Mean 15.02 7.88 .66 1.70
High-income countries
Estonia (EE) 20.21 9.00 1.40 2.40
Slovakia (SK) 23.18 10.00 .47 1.90
Portugal (PT) 26.22 10.00 1.58 2.50
Czech (CZ) 27.02 8.00 1.30 1.94
Slovenia (SI) 27.52 10.00 1.82 2.65
Israel (IL) 27.58 10.00 4.15 1.48
Korea (KR) 28.39 8.00 3.29 1.78
Trinidad & Tobago (TT) 29.12 10.00 .06 1.62
Greece (GR) 30.44 10.00 .63 1.35
New Zealand (NZ) 30.50 10.00 1.26 2.38
Japan (JP) 31.86 10.00 3.36 1.29
Spain (ES) 32.81 10.00 1.35 1.83
Italy (IT) 34.17 10.00 1.22 1.95
162
J.K.-H.Ma et al.
1 3
Table 6 (continued)
GDP per capita Composite
polity score
R&D as %
of GDP
Secondary educational
expenditures as % of
GDP
France (FR) 34.81 9.00 2.21 2.60
United Kingdom (GB) 36.37 10.00 1.75 2.60
Germany (DE) 37.12 10.00 2.73 2.26
Finland (FI) 37.55 10.00 3.75 2.76
Belgium (BE) 37.64 8.00 1.97 2.77
Denmark (DK) 39.62 10.00 3.07 2.86
Sweden (SE) 39.67 10.00 3.42 2.52
Australia (AU) 40.21 10.00 2.40 1.92
Austria (AT) 40.63 10.00 2.61 2.70
Ireland (IE) 41.88 10.00 1.63 2.09
Hong Kong (HK) 43.94 − 7.00 .77 1.21
Netherlands (NL) 44.40 10.00 1.69 2.23
United States (US) 47.00 10.00 2.82 1.97
Switzerland (CH) 49.92 10.00 2.73 2.10
Norway (NO) 56.19 10.00 1.72 2.54
Singapore (SG) 61.60 − 2.00 2.16 .75
Macao (MO) 76.85 − 7.00 .05 .92
Mean 37.81 8.20 1.98 2.06
Note GDP per capita is in thousands of 2009 purchasing power parity dollars. The classification of income
groups is based on the ITU (2011). Countries within groups are sorted by GDP per capita. All of the coun-
try-level data are from 2009. For countries with missing data on R&D as % of GDP in 2009, we use the
closest available data year: Australia (2008), Georgia (2005), Jordan (2008), Mauritius (2005), Peru (2004),
and Switzerland (2008). For countries with missing data on secondary educational expenditures as % of
GDP in 2009, we use the closest available data year: Costa Rica (2004), Georgia (2008), Greece (2005),
Japan (2008), Macao (2000), Panama (2007), Shanghai (1999), Slovenia (2003), Tunisia (2008), and Uru-
guay (2006)
163
National Income, Political Freedom, andInvestments inR&D…
1 3
References
Attewell, P. (2001). The first and second digital divides. Sociology of Education, 74(3), 252–259.
Attewell, P. (2003). Beyond the digital divide. In P. Attewell & N. Seel (Eds.), Disadvantaged Teens and
Computer Technologies (pp. 15–34). New York: Waxmann.
Attewell, P., Suazo-Garcia, B., & Battle, J. (2003). Computers and young children: Social benefit or social
problem? Social Forces, 82(1), 277–296.
Fig. 3 Effects of R&D and secondary educational expenditures on family SES slopes by country income
group: digital use for schoolwork at home vs. use of educational software at homea. Note Predicted fam-
ily SES slopes are based on the same model specifications in Table4 but limited to only 36 countries with
valid data for the two dependent variables. a Standardized composite measure based on five items: (1)
browsing the Internet for schoolwork, (2) using e-mail to communicate with other students about school-
work, (3) using e-mail to communicate with teachers and for submission of schoolwork, (4) using material
from the school’s website, and (5) checking the school’s website for announcements
164
J.K.-H.Ma et al.
1 3
Baker, D., Goesling, B., & LeTendre, G. (2002). Socioeconomic status, school quality, and national eco-
nomic development: A cross-national analysis of the “Heyneman-Loxley Effect” on mathematics and
science achievement. Comparative Education Review, 46(3), 291–312.
Becker, H. J. (2000). Who’s wired and who’s not: Children’s access to and use of computer technology. The
Future of Children, 10(2), 44–75.
Bonfadelli, H. (2002). The internet and knowledge gaps: A theoretical and empirical investigation. Euro-
pean Journal of Communication, 17(1), 65–84.
Brown, P. (2013). Education, opportunity and the prospects for social mobility. British Journal of Sociology
of Education, 34(5–6), 678–700.
Bryan, M., & Jenkins, S. (2016). Multilevel modelling of country effects: A cautionary tale. European Soci-
ological Review, 32(1), 3–22.
Buchmann, C., & Hannum, E. (2001). Education and stratification in developing countries: A review of
theories and research. Annual Review of Sociology, 27, 77–102.
Carnoy, M., & Rothstein, R. (2013). What do international tests really show about U.S. student perfor-
mance? http://www.epi.org/publi catio n/us-stude nt-perfo rmanc e-testi ng/. Accessed Oct 2015.
Chankseliani, M. (2014). Georgia: Marketization and education post-1991. In N. Ivanenko (Ed.), Education
in eastern Europe and Eurasia (pp. 277–302). London; New York: Bloomsbury Academic.
Chiu, M. M. (2010). Effects of inequality, family and school on mathematics achievement: Country and stu-
dent differences. Social Forces, 88(4), 1645–1676.
Corrales, J., & Westhoff, F. (2006). Information technology adoption and political regimes. International
Studies Quarterly, 50(4), 911–933.
Dale, R. (2005). Globalisation, knowledge economy and comparative education. Comparative Education,
41(2), 117–149.
DeBell, M., & Chapman, C. (2006). Computer and internet use by students in 2003 (NCES 2006-065).
Washington, DC: U.S. Department of Education. National Center for Education Statistics.
Decancq, K., & Schokkaert, E. (2016). Beyond GDP: Using equivalent incomes to measure well-being in
Europe. Social Indicators Research, 126(1), 21–55. https ://doi.org/10.1007/s1120 5-015-0885-x.
DiMaggio, P., & Cohen, J. (2005). Information inequality and network externalities: A comparative study of
the diffusion of television and the Internet. In V. Nee & R. Swedberg (Eds.), The economic sociology
of capitalism. Princeton: Princeton University Press.
DiMaggio, P., Hargittai, E., Celeste, C., & Shafer, S. (2004). Digital inequality: From unequal access to
differentiated use. In K. Neckerman (Ed.), Social inequality (pp. 355–400). New York: Russell Sage
Foundation.
Drori, G. S. (2006). Global e-litism: Digital technology, social inequality, and transnationality. New York,
NY: Worth Publishers.
Drori, G. S. (2010). Globalization and technology divides: Bifurcation of policy between the “digital
divide” and the “innovation divide”. Sociological Inquiry, 80(1), 63–91.
Dutton, W. H., Gillett, S. E., McKnight, L. W., & Peltu, M. (2004). Bridging broadband Internet divides:
Reconfiguring access to enhance communicative power. Journal of Information Technology, 19(1),
28–38.
Education Services Australia. (2012). National digital learning resources network. http://www.ndlrn .edu.au/
defau lt.asp. Accessed July 2015.
Elwert, F., & Winship, C. (2014). Endogenous selection bias: The problem of conditioning on a collider
variable. Annual Review of Sociology, 40, 31–53.
Erichsen, E. A., & Salajan, F. D. (2014). A comparative analysis of e-learning policy formulation in the
European Union and the United States: Discursive convergence and divergence. Comparative Educa-
tion Review, 58(1), 135–165.
Fuchs, T., & Wossmann, L. (2004). Computers and student learning: Bivariate and multivariate evidence
on the availability and use of computers at home and at school. Brussels Economic Review, 47(3–4),
359–386.
Gamoran, A. (2001). American schooling and educational inequality: A forecast for the 21st century. Sociol-
ogy of Education, 74, 135–153.
Ganzeboom, H. B. G., De Graaf, P. M., & Treiman, D. J. (1992). A standard international socio-economic
index of occupational status. Social Science Research, 21(1), 1–56.
Goode, J. (2010). The digital identity divide: How technology knowledge impacts college students. New
Media and Society, 12(3), 497–513.
Guillén, M. F., & Suárez, S. L. (2005). Explaining the global digital divide: Economic, political and socio-
logical drivers of cross-national Internet use. Social Forces, 84(2), 681–708.
Hannum, E., & Buchmann, C. (2005). Global educational expansion and socio-economic development: An
assessment of findings from the social sciences. World Development, 33(3), 333–354.
165
National Income, Political Freedom, andInvestments inR&D…
1 3
Hargittai, E., & Hinnant, A. (2008). Digital inequality: Differences in young adults’ use of the Internet.
Communication Research, 35(5), 602–621.
Hargittai, E., & Shafer, S. (2006). Differences in actual and perceived online skills: The role of gender.
Social Science Quarterly, 87(2), 432–448.
Heisig, J. (2011). Who does more housework: Rich or poor? A comparison of 33 countries. American
Sociological Review, 76(1), 74–99.
Heisig, J., Schaeffer, M., & Giesecke, J. (2017). The costs of simplicity: Why multilevel models may
benefit from accounting for cross-cluster differences in the effects of controls. American Sociologi-
cal Review, 82(4), 796–827.
Heyneman, S. P., & Loxley, W. A. (1983). The effect of primary-school quality on academic achievement
across twenty-nine high- and low-income countries. American Journal of Sociology, 88(6), 1162–1194.
Hilbert, M. (2010). When is cheap, cheap enough to bridge the digital divide? Modeling income related
structural challenges of technology diffusion in Latin America. World Development, 38(5), 756–770.
Hill, P. (2010). Learning as we go: Why school choice is worth the wait. Stanford, CA: Hoover Institution
Press.
Imhof, M., Vollmeyer, R., & Beierlein, C. (2007). Computer use and the gender gap: The issue of access,
use, motivation, and performance. Computers in Human Behavior, 23(6), 2823–2837.
ITU. (2011). Measuring the information society. Geneva: ITU. http://www.itu.int/en/ITU-D/Stati stics /Pages
/publi catio ns/mis20 11.aspx. Accessed Dec 2013.
Juárez, F., & Gayet, C. (2014). Transitions to adulthood in developing countries. Annual Review of Sociol-
ogy, 40, 521–538.
Kim, S.-H. (2008). Testing the knowledge gap hypothesis in South Korea: Traditional news media, the Inter-
net, and political learning. International Journal of Public Opinion Research, 20(2), 193–210.
Leu, D. J., Forzani, E., Rhoads, C., Maykel, C., Kennedy, C., & Timbrell, N. (2014). The new literacies of
online research and comprehension: Rethinking the reading achievement gap. Reading Research Quar-
terly, 50(1), 37–59.
Leung, L., & Lee, P. S. N. (2012). Impact of Internet literacy, Internet addiction symptoms, and Internet
activities on academic performance. Social Science Computer Review, 30(4), 403–418.
Livingstone, S., & Helsper, E. J. (2008). Parental mediation and children’s internet use. Journal of Broad-
casting and Electronic Media, 52(4), 581–599.
Marshall, M. G., Gurr, T. R., & Jaggers, K. (2010). Polity IV project: Political regime characteristics and
transitions, 1800–2009. http://www.syste micpe ace.org/polit y/polit y4.htm. Accessed Nov 2011.
Mayer, S. E. (2001). How did the increase in economic inequality between 1970 and 1990 affect children’s
educational attainment? American Journal of Sociology, 107(1), 1–32.
Milioni, D., Doudaki, V., & Demertzis, N. (2014). Youth, ethnicity, and a ‘reverse digital divide’: A study of
Internet use in a divided country. Convergence: The International Journal of Research into New Media
Technologies, 20(3), 316–336.
Natriello, G. (2001). Bridging the second digital divide: What can sociologists of education contribute?
Sociology of Education, 74(3), 260–265.
Norris, P. (2001). Digital divide? Civic engagement, information poverty, and the Internet worldwide. New
York: Cambridge University Press.
Notten, N., Peter, J., Kraaykamp, G., & Valkenburg, P. M. (2009). Research note: Digital divide across
borders—A aross-national study of adolescents’ use of digital technologies. European Sociological
Review, 25(5), 551–560.
OECD. (2011). PISA 2009 results: Students on line: Digital technologies and performance (volume VI).
Paris: OECD Publishing. https ://doi.org/10.1787/97892 64112 995-en.
OECD. (2012). PISA 2009 technical report. PISA, OECD Publishing. http://dx.doi.org/10.1787/97892
64167 872-en.
OECD. (2015). Students, computers and learning: Making the connection. Paris: PISA, OECD Publishing.
https ://doi.org/10.1787/97892 64239 555-en.
Park, H., & Kyei, P. (2011). Literacy gaps by educational attainment: A cross-national analysis. Social
Forces, 89(3), 879–904.
Peter, J., & Valkenburg, P. M. (2006). Adolescents’ internet use: Testing the “disappearing digital divide”
versus the “emerging digital differentiation” approach. Poetics, 34(4–5), 293–305.
Powell, W. W., & Snellman, K. (2004). The knowledge economy. Annual Review of Sociology, 30, 199–220.
Rabe-Hesketh, S., & Skrondal, A. (2008). Multilevel and longitudinal modeling using Stata. College Sta-
tion, TX: Stata Press.
Raftery, A. E., & Hout, M. (1993). Maximally maintained inequality: Expansion, reform, and opportunity in
Irish education, 1921–75. Sociology of Education, 66(1), 41–62.
166
J.K.-H.Ma et al.
1 3
Robinson, L. (2014). Freeways, detours, and dead ends: Search journeys among disadvantaged youth. New
Media and Society, 16(2), 234–251.
Robison, K. K., & Crenshaw, E. M. (2010). Reevaluating the global digital divide: Socio-demographic and
conflict barriers to the Internet revolution. Sociological Inquiry, 80(1), 34–62.
Royston, P., Carlin, J., & White, I. (2009). Multiple imputation of missing values: New features for mim.
Stata Journal, 9(2), 252–264.
Ruiter, S., & van Tubergen, F. (2009). Religious attendance in cross-national perspective: A multilevel anal-
ysis of 60 countries. American Journal of Sociology, 115(3), 863–895.
Smith, J., Skrbis, Z., & Western, M. (2013). Beneath the ‘digital native’ myth: Understanding young Aus-
tralians’ online time use. Journal of Sociology, 49(1), 97–118.
Snow, C. (2002). Reading for understanding: Toward an R&D program in reading comprehension. Santa
Monica, CA: RAND Corporation.
Spring, J. (2008). Research on globalization and education. Review of Educational Research, 78(2),
330–363.
UNESCO. (2015). Education for all 2000–2015: Achievements and challenges. Paris: UNESCO. https ://
en.unesc o.org/gem-repor t/repor t/2015/educa tion-all-2000-2015-achie vemen ts-and-chall enges #sthas
h.bfFmE FAV.dpbs. Accessed July 2015.
UNESCO. (2016). Information and communication technologies for development (ICT4D). http://www.
ict4d c.org/. Accessed July 2015.
UNU-WIDER. (2008). World income inequality database, version 2.0c, May 2008. http://www.wider .unu.
edu/resea rch/Datab ase/en_GB/wiid/. Accessed Jan 2013.
U.S. Department of Education. (1996). Getting America’s students ready for the 21st century: Meeting
the technology literacy challenge. A report to the nation on technology and education. http://eric.
ed.gov/?id=ED398 899. Accessed June 2015.
van Deursen, A., & van Dijk, J. (2014). The digital divide shifts to differences in usage. New Media and
Society, 16(3), 507–526.
Vegas, E., & Coffin, C. (2015). When education expenditure matters: An empirical analysis of recent inter-
national data. Comparative Education Review, 59(2), 289–304.
Worden, E. A. (2014). Moldova: Challenges and opportunities. In N. Ivanenko (Ed.), Education in Eastern
Europe and Eurasia (pp. 47–64). London: Bloomsbury Academic.
World Bank. (2015a). World development indicators. http://datab ank.world bank.org/data/downl oad/archi ve/
WDI_excel _2015_01.zip. Accessed Apr 2016.
World Bank. (2015b). Education statistics. http://data.world bank.org/data-catal og/ed-stats . Accessed Oct
2015.
World Bank. (2015c). Why use GNI per capita to classify economies into income groupings? https ://datah
elpde sk.world bank.org/knowl edgeb ase/artic les/37883 1-why-use-gni-per-capit a-to-class ify-econo mies-
into. Accessed Feb 2015.
- A preview of this full-text is provided by Springer Nature.
- Learn more
Preview content only
Content available from Social Indicators Research
This content is subject to copyright. Terms and conditions apply.
