Journal of Global Information Technology Management

ISSN: 1097-198X (Print) 2333-6846 (Online) Journal homepage: http://www.tandfonline.com/loi/ugit20

Exploring the Predictors of the International

Digital Divide

Maria Skaletsky, Robert D. Galliers, Dominique Haughton & Olumayokun

Soremekun

To cite this article: Maria Skaletsky, Robert D. Galliers, Dominique Haughton & Olumayokun
Soremekun (2016) Exploring the Predictors of the International Digital Divide, Journal of Global
Information Technology Management, 19:1, 44-67, DOI: 10.1080/1097198X.2016.1134171

To link to this article: http://dx.doi.org/10.1080/1097198X.2016.1134171

Published online: 25 Feb 2016.

Submit your article to this journal

Article views: 346

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at

http://www.tandfonline.com/action/journalInformation?journalCode=ugit20

Download by: [UNIVERSITY OF BRIGHTON] Date: 05 March 2017, At: 03:18

JOURNAL OF GLOBAL INFORMATION TECHNOLOGY MANAGEMENT
2016, VOL. 19, NO. 1, 4467
http://dx.doi.org/10.1080/1097198X.2016.1134171

Exploring the Predictors of the International Digital Divide

Maria Skaletskya, Robert D. Galliersb, Dominique Haughtonc, and Olumayokun Soremekund
a
Academic Technology Center, Bentley University, Waltham, Massachusetts, USA; bDepartments of Information &
Process Management and Sociology, Bentley University, Waltham, Massachusetts, USA, and School of Business &
Economics, Loughborough University, Loughborough, Leicestershire, UK; cDepartment of Mathematical Sciences and
Global Studies, Bentley University, Waltham, Massachusetts, USA, Paris University, Paris, France, and Toulouse
University, Toulouse, France; dOLX, Lagos, Nigeria

ABSTRACT KEYWORDS
The digital divide, while attracting considerable research and political atten- Global digital divide;
tion since the introduction of the Internet, remains an apparently intract- internet use; mobile phone
able issue for parts of the developing world. While most prior research subscriptions; TreeNet
focuses on national income as the most significant predictor of the digital
divide and thus leaves little room for intervention when financial resources
are scarce, the current study identifies institutional stability and infrastruc-
ture development as being of similar importance, thereby demonstrating a
route out of the otherwise apparent impasse. In this study, we employ
TreeNet, a software tool based on the gradient descent algorithm, which
has not been previously used in digital divide studies. This tool allows for
the determination of the relative importance of predictors of digital devel-
opment and the importance of predictors that propel countries to top or
bottom levels of digital development. The authors also uncover important
interactions among predictors of digital development. Separate models
were built for the level of Internet use and for the number of mobile
phone subscriptions at a country level and these models were found to
be different. As a result, the authors were able to determine critical thresh-
olds of infrastructure at which the level of digital development drastically
increases.

Introduction
The invention of the World Wide Web in 1989 (Norris, 2001) has had an immense effect on the way
information is transferred, business is conducted, and the way individuals communicate with each
other and the world. The emergence of the Internet has enhanced the lives and economies of many
but widened the divide within and between countries (Chen & Wellman, 2004; Dewan, Ganley, &
Kraemer, 2005). Information systems as a discipline has a role to play in helping to deal with such
key societal issues as this (Desouza, El-Sawy, Galliers, Loebbecke, & Watson, 2006, 2007; Lu, 2001).
Understanding underlying issues associated with the digital divide (DD) is important in order to be
able to identify the policies that would help in ameliorating the associated injustices and inequities.
The importance of bridging the divide is magnified by arguments indicating that an increase in the
use of information and communication technology (ICT) may help developing countries to leap-
frog some developmental stages and to significantly increase productivity, bypassing the traditional
steps of the accumulation of investments and human capital (Steinmueller, 2001). The DD is a policy
issue, and policies cannot be effective without an extensive understanding of the phenomena
underlying the issue.

CONTACT Maria Skaletsky mskaletsky@bentley.edu Academic Technology Center, Bentley University, 175 Forest St.,
Waltham, MA 02452, USA.
Color versions of one or more of the figures in the article can be found online at www.tandfonline.com/UGIT.
2016 Maria Skaletsky, Robert D. Galliers, Dominique Haughton, and Olumayokun Soremekun
 JOURNAL OF GLOBAL INFORMATION TECHNOLOGY MANAGEMENT 45

A prerequisite in developing effective policies is the need for policymakers to know which
predictors of the global DD are responsible for differentiating between countries with high levels
of economic and digital development and those with low levels of development. While numerous
studies address the issue of the global DD, there is still no clear understanding of which predictors
are more important than others. Many studies concentrate on identifying determinants of the divide;
however, they provide only general information about the significance level of these predictors and
do not account for the relative importance of these determinants and the complexity of relationships
between them and digital development. Most studies assume linearity in relationships between the
predictors and the level of digital development and do not account for possible interactions among
predictors (Andonova, 2006; Bagchi, 2005; Bagchi, Hart, & Peterson, 2004; Bagchi & Udo, 2007;
Beilock & Dimitrova, 2003; Billon, Marco, & Lera-Lopez, 2009, 2010). While identifying significant
predictors of the DD is important, more information is needed to understand which factors are more
important than others and the way predictors interact. Identifying the relative importance of the
predictors of the DD and the complexity of their relationships with digital development may thus be
of help to policy makers in addressing problems associated with the divide, given their particular
countrys circumstances and context.
In this study, we build on previous literature (Billon et al., 2010; Chinn & Fairlie, 2007;
Deichmann et al., 2006) and provide a detailed analysis of commonly used predictors of the global
DD. TreeNet, a software tool based on the gradient descent algorithm (Friedman, 2001), was
employed, which has not been previously used in DD studies. TreeNet allows detailed information
about the nature of the relationship between digital development and its predictors to be obtained.
The main contributions of this study are the following: (1) The authors determined the critical
levels of infrastructure and other predictors of the DD at which the level of digital development
drastically increases by uncovering non-linear relationships between digital development and its
predictors. (2) Separate models were built for the level of Internet and mobile phone penetration and
different results for both models were obtained, which provides valuable information about the
difference in critical levels needed to jump start these different technologies. (3) There were
important interactions discovered among the predictors, which allows for obtaining a more detailed
picture of the conditions associated with an increase in rates of use of Internet and mobile phones.
(4) The importance of predictors of digital development was evaluated, identifying those that can be
leveraged by poorer countries. The authors also evaluate the importance of the predictors that propel
countries to top or bottom levels of digital development.
This article is structured as follows; First, the analytical techniques used in prior research were
considered as a basis for the selection of the variables to be studied and with a view to developing a
model of the predictors of the DD. The authors are then in a position to discuss an appropriate
research method before presenting the findings of the current study. The article is concluded with a
discussion on the theoretical and practical contributions.

Analytical Techniques Used in Prior Research

Several past studies have addressed the relationships among the predictors of the global DD and
digital development. Dewan et al. (2005) used quintile regression to assess the difference in the effect
of the predictors of the DD at different levels of personal computer (PC) and Internet use. Quintile
regression is an extension of traditional ordinary least square (OLS) regression (Dismuke &
Lindrooth, 2006; Koenker & Hallock, 2001). While OLS regression estimates conditional mean
functions, quintile regression estimates conditional quintile functions and, therefore, allows for
examining the impact of the predictors at different levels of the dependent variable. The sample
used by Dewan and colleagues included 40 countries for the years 19852001. The important
contribution of this study is that the authors were able to identify those factors that are widening
the divide and those factors that contribute to closing the divide. They determine that the feedback
effect between levels of income per capita and information technology (IT) penetration is the
46 M. SKALETSKY ET AL.

primary driver of the DD: the higher the level of IT penetration, the higher the positive association
between income per capita and IT penetration. Another important conclusion is that developing
countries benefit more than developed countries from improvement in education, lowering costs of
infrastructure and increasing participation in the global economy. One important limitation of this
study, however, is that African countries were excluded from the sample due to lack of data. As a
result, the analysis excluded many countries with the lowest levels of IT penetration. Another
limitation is the omission of a measure of institutional quality, which was found to be important
in Chinn and Fairlie (2007), among others.
Deichmann et al. (2006) employ multivariate adaptive regression splines (MARS) to advance
understanding of the relationships among the variables that are commonly used as predictors of the
divide and digital development. This methodology allows for determining the interactions and non-
linearities among the variables used in a multiple regression, which in turn provides more detailed
and precise information about these relationships (Friedman, 1991). The authors used three cate-
gories of variables in their study: socio-economic, ICT price-related and ICT policy and infrastruc-
ture-related variables. The study considered 160 countries for the year 2003 and confirmed that most
relationships are not linear and that the interactions among variables have an effect on the level of
digital development. For example, the level of infrastructure development must be relatively high
before an additional improvement in it has a positive effect on the proportion of Internet users.
Similarly, an increase in the number of years of education is associated with a higher proportion of
Internet use in the population, but only after reaching a 10-year point. This study revealed more
complicated relationships among variables than had been previously shown. This study was
advanced by using more recent data and by including separate analyses for groups of countries
with different levels of digital development, thus providing an additional level of detail about these
relationships.
Chinn and Fairlie (2007) used a wide range of variables, both socio-economic and demographic,
in their study of the factors determining Internet penetration in both advanced and emerging
countries. They examined the relative importance of factors, such as income, human capital,
infrastructure, cost of use, and regulatory quality. The study confirmed the importance of income
per capita as the main determinant of Internet penetration and computer use, as well as the
importance of the level of infrastructure, measured by the number of mail telephone lines per 100
people and electric power consumption. They also determined that the price of ICT use, as measured
by the cost of monthly subscription and the cost of telephone calls, is not a significant predictor and
that regulatory quality is an important determinant of computer and Internet use. The level of
education was found to be a significant but surprisingly unimportant predictor of computer and
Internet use. Age was found to be negatively correlated with computer penetration. To determine the
relative importance of the factors that account for differences between groups of countries, the
authors use the Blinder-Oaxaca technique (Jann, 2008) to decompose a gap between groups of
countries. This technique allows for analyzing two groups of countries at a time. Chinn and Fairlie
calculate the decomposition for computer and Internet penetration rate gaps between the United
States and different regions of the world one at a time. They found income to be the most important
determinant of the difference in PC and Internet penetration between the United States and all other
regions. The number of telephone lines per capita and electricity consumption are also found to be
important determinants. A limitation of this approach is that countries in the same geographical
region are not necessarily homogeneous in their development. Therefore, analyzing differences
among geographical regions may provide skewed results. Another limitation is the potential issue
of multi-collinearity, common to linear regressions.
Bagchi et al. (2004) investigate associations between cultural dimensions of a country and the
adoption of IT products and, very interestingly, find that such associations are essentially supported
by data, albeit with the limitation that the sample is necessarily limited to a set of countries where
data on cultural dimensions are available, and that linear dependences are assumed between the
target and predictors.
 JOURNAL OF GLOBAL INFORMATION TECHNOLOGY MANAGEMENT 47

Pick and Azari (2011) design and test an exploratory model that evaluates the effects of govern-
ment support and openness, business and technology investment, socioeconomic factors on tech-
nology use for a set of 110 counties, both developed and developing. They employ a structural
equation model to test a conceptual model separately for a set of developed and developing
countries. They identify a significant indirect path from the construct of government support and
openness to the technology utilization construct through the socio-economic level construct for the
set of developed countries. The study also identifies a significant link from the socio-economic
construct and from the business and technology investment construct to the technology utilization
construct for developing countries. One of the limitations of this study is that countries are split into
developed and developing categories based on the World Bank classification. This group of countries
is very diverse in their technological development. For instance, Saudi Arabia is included in the
developing countries category but has one of the highest rates of mobile phone use. Also, the
inclusion of different technologies into a combined technology utilization factor may reduce the
ability to understand how different technologies are adopted. In the current study, separate models
for countries in the top and bottom quintiles were built based on their Internet and mobile phone
use separately. The authors expected to find different results for the Internet adoption model and for
the mobile phone adoption model. The authors are also able to account for non-linearities among
the predictors of technology adoption.
Billon et al. (2010) employ an OLS regression model for a set of 142 developing and developed
countries to determine which predictors of digital development are significant for countries with
different digitalization levels. They find that different variables explain different digitalization levels.
Gross domestic product (GDP) is found to be important in predicting the level of digitalization for
middle income countries, while the quality of regulation is most important for highly digitalized
countries. Infrastructure is found to be the most important predictor of digital development for
countries with the lowest levels of digitalness. The authors added to this study and build separate
models for Internet and mobile phone penetration instead of using a digitalization index. This allows
for the differences between the determinants of adoption of these different technologies to be found.
A different methodology was employed, which in addition to the assessment of importance of
predictors of levels of digitalness, allows for finding critical levels for each important predictor at
which levels of Internet and mobile phone use drastically increase and also allows uncovering of
interesting interactions among the predictors. The differences and similarities with previous studies
were summarize in Table 8.
The authors finally note that a recent effort to address non-linear dependencies between target
and predictors in IT studies was described in Bankole, Osei-Bryson, and Brown (2013). The article
considers the association between ICT investment and human development and employs MARS
models (Friedman, 1991).
While the studies described above provide valuable insight regarding the predictors of the global
DD, the current study builds on this work and provides additional insights into the relationship
between digital development and its predictors, such as critical levels of predictors of Internet and
mobile phone penetration, interesting interactions among the predictors and the importance of
predictors of countries being placed into the top and the bottom levels of digital development. The
studies above guided the selection of variables that are described in the next section.

Variable Selection
Dependent VariableMeasure of Digitalness
In order to build a model to evaluate the importance of the predictors of the DD, the authors needed
to decide on a way to operationalize the measure of digitalness that would be used as a dependent
variable in the models. Several different approaches were identified in the existing literature. Some
studies use a single measure of digital development, such as the number of Internet users per 1,000
48 M. SKALETSKY ET AL.

people (e.g., Dasgupta, Lall, & Wheeler, 2005; Hargittai, 1999; Kiiski & Pohjola, 2002). Some use
several dependent variables and construct separate models for each measure. For instance, Chinn
and Fairlie (2007) use two dependent variables, the number of PCs per 100 people and the number
of Internet users per 100 people. They find that income, electricity consumption, regulatory quality,
and proportion of urban population are significant predictors of both Internet and PC penetration.
Dewan et al. (2005) use three separate measures of digital development: the number of mainframe
computers, the number of PCs, and the number of Internet users per capita and per unit of GDP.
They find that for all technologies, IT penetration is positively associated with national income, level
of education, and the size of trade and negatively associated with the proportion of urban population
and telecommunication costs. They also find some indicators that are significant in predicting one of
the three measures of digital development but not the others. For instance, the proportion of urban
population has almost no effect on Internet diffusion but impacts the PC and mainframe computer
penetration.
Another option for measuring digitalness is to construct an index that would include several
measures at the same time. At the global level, Sciadas (2005) uses an Infostate index, which
combines measures of Infodensity, which in turn includes eight measures of levels of infrastructure
and five measures of educational levels, and Info-use, which includes four measures of digital
development (measured by PC, TV, and phone ownership) and three measures of intensity of use
(measured by the number of outgoing and incoming international phone calls and the number of
Internet users). First, a reference index is calculated for a reference year 2001 for a reference country,
which has average values over all countries. Indicators are calculated using different formula,
depending on the nature of the variables involved. For example, for the Internet, the index is
calculated as follows:
 secure servers 
1  100
I Internet hosts
population
Once all indicators are calculated, they are adjusted by the value of the indicator or a reference
country for a reference year. Then, for each sub-index a product of involved indicators is calculated
and sub-indexes in turn are combined into one index, Infostate. The index is then used to compare
the levels of development among countries.
Bagchi (2005) uses factor analysis to extract a single factor from four different technologies: main
line phones, cell phones, PCs, and internet users per 1,000 of population. This study analyzes
correlations between this single factor and several predictors of the DD and conducts a regression
analysis with this factor as a target.
Another example of an index used to evaluate the DD is the Technology Achievement Index
(2011). The index includes measures of technology creation (patents per capita and license fees from
abroad per capita), measures of diffusion of recent innovations (Internet and export of high
technology products), diffusion of old innovations (number of telephone lines and electricity
consumption) and measures of human skills (average number of years of education and the rate
of enrollment into science, mathematics, and engineering programs). Each indicator included in the
index is weighted equally. While allowing for an easy comparison among countries, the use of such
extensive indexes does not allow for discovering the importance of the measures included into their
calculation.
The current dataset includes three measures of digital development. Two measures consistent
with past literature were includedthe number of PCs per 100 people and the number of Internet
users per 100 peopleand we add a variable measuring the number of mobile phone subscriptions
per 100 people. The measure of cell phone adoption was also included in the analyses because of its
exponential increase in recent years, particularly in emerging economies. This measure was not part
of the analysis in any of the prior studies reviewed earlier in this article. In order to decide on the
dependent variable, correlations among these variables (see Table 1) were explored.
 JOURNAL OF GLOBAL INFORMATION TECHNOLOGY MANAGEMENT 49

Table 1. Correlation Between the Three Measures of Digital Development.

Number of PC per Number of Internet Users per Number of Mobile Subscriptions
Name of the Indicator 100 People 100 People per 100 People
Number of PC per 100 people 1 0.936** 0.652**
Number of Internet users per 100 0.936** 1 0.664**
people
Number of mobile subscriptions 0.652** 0.664** 1
per 100 people
Note: **Correlation is significant at the 0.01 level (2-tailed).

Based on the high level of correlation between measures of the Internet and PC penetration
(0.94), the authors chose to select one of these variablesnumber of Internet users per 100 people
as a target variable. Correlations among mobile phone use and Internet use (66%) and among mobile
phone use and PC ownership (65%) are approximately the same. While these correlations are
relatively high, it was argued that the best approach is to construct separate models for Internet/
PC penetration and for mobile phone penetration, given that some of the information specific to
mobile phone use might be lost in a combined index. Also, given the increased preponderance of
mobile phone use in emerging economies in recent years, the authors chose to build separate models
for Internet/PC use and for mobile phone use.

Independent Variables
The choice of independent variables is guided by existing literature. In particular, the studies by
Chinn and Fairlie (2007), Dewan et al. (2005) and Billon et al. (2010) were relied on Figure 1

Figure 1. Framework for selecting the predictors of the global digital divide.
50 M. SKALETSKY ET AL.

Table 2. Indicators Used in the Study.

Indicator Description Source Group
Internet Number of Internet users per 100 people World Bank (2011) Digital
Mobile Number of mobile subscribers per 100 people World Bank (2011) development
Income GNI per capita in international ppp dollars World Bank (2011); The World Factbook (n.d.)
Gini Average Gini index for reported years World Bank (2011); The World Factbook (n.d.) Economic
Trade Trade in goods as a percentage of GDP World Bank (2011)
Maintel Number of main telephone lines per 100 World Bank (2011)
Costcall Cost of 3-minute local call ($PPP) International Telecommunication Union (2012) Infrastructure
Electric Electricity consumption kwh/capita World Bank (2011); The World Factbook (n.d.)
P1564 Percentage of population age 1564 World Bank (2011)
P65plus Percentage of population 65 and older World Bank (2011) Demographic
PSchool Primary school enrollment World Bank (2011)
SSchool Secondary school enrollment World Bank (2011)
Urban Urban population as a percent of total World Bank (2011)
Risk Country risk rating IHS Global Insight (2009) Risk

demonstrates the framework for selecting the predictors of the global DD, similar to those used in
these studies. Each of these studies includes variables describing demographic characteristics of
technology consumers, level of economic development of the countries, levels of institutional
stability, and infrastructure development. A detailed description of the theoretical foundation for
the choice of variables can be found in Billon et al. (2010). The authors build on their efforts while
expanding on them to provide new insights. The list of all variables included in this study is
presented in Table 2.
Several indicators representing economic measures were included. GDP per capita is a national
income measure, which has been found to be an important predictor of the DD in past studies
(Chinn & Fairlie, 2007; Deichmann et al., 2006; Dewan et al., 2005). Consistent with past literature,
the authors expected to find income to be an important predictor of global digitalness. A measure of
intra-country inequalitythe Gini index (Lerman & Yitzhaki, 1984)and trade in goods was also
included as a percentage of GDP, which were also found to be significant in past studies.
Several demographic indicators were also included: percentage of urban population, percentage of
younger population, and two measures of educationwhich were found to be positively associated
with the PC and Internet penetration in the past. Another demographic indicatorpercentage of
older populationwas found to be negatively correlated with PC and Internet use. Several measures
of infrastructure were included, and the authors expected to find these measures to be important
predictors of digital development.
The last indicator included in the model is the Information Handling Services (IHS) country risk
rating,1 which incorporates economic, political, legal, tax, operational, and security risk ratings. This
variable is a more comprehensive risk index than was used previously in the Deichmann et al. (2006)
study and is a measure of the overall institutional stability in a country. Institutional quality has long
been considered to be an important factor in national and regional economic development.
Institutional quality is found to be a prerequisite for trade openness (Menzie et al., 2006), which
has been found to be an important antecedent of foreign direct investments (FDI) inflow in
transition economies (Bevan, Estrin, & Meyer, 2004) and also an important predictor of regional
economic development in China (Hasan, Wachtel, & Zhou, 2009). A measure of institutional quality
was found to be important in differentiating between U.S. and Middle Eastern countries in the level
of Internet penetration (Chinn & Ito, 2006). A measure of institutional quality was also used in the
study by Martinez and Williams (2010) in evaluating competing theories explaining the development
of e-commerce, which was operationalized as ICT adoption. Economic institutional theory predicts
that effective and strong national institutions foster trust and an increase in the level of e-commerce
and ICT adoption. The authors find the risk measure to be a significant predictor of ICT adoption.
Billon et al. (2010) find the quality of regulation to be an important predictor of digital development
 JOURNAL OF GLOBAL INFORMATION TECHNOLOGY MANAGEMENT 51

for a group of countries with high levels of digital development. The authors expected to find this
measure to be an important predictor of the global DD.
Missing data in the current data set were imputed using the multiple imputation function in IBM
SPSS 19. All variables with missing values were regressed on all other predictors except for the
measures of digitalness (number of Internet users and mobile phone subscribers).
Thus, consistent with prior research, the current dataset includes variables describing demo-
graphic characteristics of the population, the level of economic development of the countries, levels
of infrastructure development, and a measure of institutional stability. The methodology chosen for
the analyses and its advantages over methods used in prior studies is described in the next section.

MethodologyThe TreeNet Algorithm

TreeNet was employed, a software tool based on the gradient descent algorithm (Friedman, 2001).
The objective of the TreeNet analysis is to detect non-linear relationships that exist between the
target variable and predictors with a view to uncovering interactions among predictors and to
determine the relative importance of the predictors. In the case of the DD, a non-linear relationship
would mean that the level of Internet use does not gradually increase at a constant rate as one of its
predictors gradually increases, but rather it changes at a different rate at different levels of the
predictor. This functionality allowed for critical levels of predictors to be pinpointed at which the
level of Internet use and Mobile phone use drastically increase. While this does not establish cause
and effect relationships, it was posited that it provides important policy implications.
Another advantage of this algorithm is that it allows for finding interactions in the effects of
predictors. While interactions are identified by other softwareparticularly MARS, as used by
Deichmann et al. (2006)partial effects graphs are not available in their study. The importance of
predictors was also able to be determined, which propels countries to top or bottom levels of digital
development. The closest method that allows doing so is quintile regression, as used by Dewan et al.
(2005). The advantage of TreeNet over quintile regression is that it is a non-parametric technique and
does not require data assumptions necessary for a regression, such as independence and normality of
residuals and homoscedasticity. It is also not impacted by multi-collinearity issues. Another advantage
of TreeNet is that quintile regression assumes linear relationships between the dependent variable and
its predictors. Post hoc, the current results clearly show that these relationships are not linear.
A further advantage of TreeNet is the ranking of the importance of predictors in their effect on
the dependent variable. TreeNet is not unique in calculating the importance level: Comparison of
standardized coefficients in a regular linear regression also allows for an evaluation of the impor-
tance of predictors. However, as discussed previously, regression methodology has many limitations
that are avoided by TreeNet. Thus, while regression analyses used in past studies (e.g., Chinn &
Fairlie, 2007; Dewan et al., 2005) provides useful results regarding the level of significance of each of
the predictors, and general information about the effects of independent variables on the dependent
variable, TreeNet allows for obtaining much more detailed information about these relationships.
TreeNet is a relatively new tool; it was created in 1999 by Stanford University Professor Jerome
Friedman and is built on the methodology known as stochastic gradient boosting (Friedman, 2001).
TreeNet gradually builds a high numberpotentially thousandsof small decision trees, the sum of which
converge to the resulting model through an error-correction strategy. Each small tree typically has
approximately six terminal nodes. Each added tree acts as a boost to the performance of the previous
tree. To avoid over-fitting, the number of trees is controlled by an evaluation of performance on the test
sample.
A decision tree can be viewed as a function of the predictors (referred to as a step-function) that is
constant on each terminal node. The following example explains the way TreeNet works. In this
example, the dependent variable Y is continuous. A simple decision tree built with two predictors X1
and X2 (Figure 2) generates the following step-function:
52 M. SKALETSKY ET AL.

Figure 2. Simple decision tree.

Tree X1 ; X2 Y1 ; if X1  a
Tree X1 ; X2 Y2 ; if X1 >a and X2  b
Tree X1 ; X2 Y3 ; if X1 >a and X2 >b
where Y1, Y2, and Y3 represent the means of the target variable on the terminal nodes 1, 2, and 3,
respectively. The TreeNet procedure uses a sum of step-functions to link the target variable Y to a set
of predictors X, where each step function corresponds to a tree with a small number of nodes.
The model is similar to a long series expansion, where a sum of factors becomes progressively
more accurate with each step. It can be written as following:
F X F0 1 Tree1 X 2 Tree2 X . . . m Treem X
where each Tree1 is a small tree. Each additional tree provides a boost to the previous model. In
this formula, the i and the splits of the small trees are determined by a gradient descent algorithm
(Friedman, 2001). The goal of this algorithm is to minimize the mean square error of the approx-
imation with each additional tree. When the target variable is categorical, a different error function is
used instead of the mean square error (Friedman, 2001).
A more detailed description of the TreeNet methodology can be found in Friedman (2001).
TreeNet has been used in past literature. One example of TreeNet use is the study by Eshghi et al.
(2011), where the algorithm was used to determine the most important factors affecting graduate
students attrition rate. In the next section, the findings arising from the models of Internet use and
mobile phone penetration in the current study are presented and discussed.

Findings
Predictors of Internet Penetration
The first model explores the importance of predictors of Internet use for all countries included in
the dataset. The three most important predictors, with an importance level over 50%, are presented
in Table 3. The importance score is a sum of improvements for all splits of all trees associated with
a variable. The importance scores are then rescaled so that the most important variable receives a
score of 100, and the importance of all other variables is calculated relative to the most important
variable.
 JOURNAL OF GLOBAL INFORMATION TECHNOLOGY MANAGEMENT 53

Table 3. Variable Importance for Internet Penetration.

Variables Relative Importance
GNI per capita 100%
Risk 99.49%
Population aged 65 and over 54.98%

Figure 3. Partial effects of predictors on the Internet use rate.

The income per capita and risk variables (IHS country risk rating, which incorporates economic,
political, legal, tax, operational, and security risk ratings) are determined to be the most important
predictors of the Internet penetration rate, with almost the same level of importance. The only other
variable with an importance level higher than 50% is the proportion of population with ages 65 and
older. Figure 3 shows partial effects of the predictors with the highest level of importance on the
Internet use rate, after taking into consideration the effects of all other variables. Figure 3 shows that
the level of Internet use increases sharply when gross national income (GNI) per capita reaches
$14,000 and levels off when GNI reaches approximately $20,000. After that point, GNI per capita
does not have an effect on the Internet use rate.
The risk level does not have an effect on the level of Internet use until the value of risk reaches
approximately 2, at which point the level of Internet use drops sharply and then stabilizes again at a
lower level when the risk value reaches three.

Predictors of Internet Quintile Membership

In the second model, we create a variable indicating to which quintile a country belongs based on the
rate of Internet use. Quintile #5 includes countries with the highest level of Internet use while
quintile #1 includes countries with the lowest level of Internet use. Examples of countries in quintile
#5 include Finland, Hong Kong, Ireland, and Japan, while quintile #1 has a high proportion of
African countries. Examples of countries in quintile #1 include India, Bangladesh, Botswana, the
Central African Republic, and Ethiopia. Quintile membership was used as a dependent variable in
this model. The authors were mostly interested in the top and the bottom quintile and only present
results for these two quintiles. Descriptive statistics of the rate of Internet use for different quintiles
are shown in Table 4. The full list of countries in the top and the bottom quintiles is provided in
Appendix 1.
TreeNet allows for obtaining the list and the rank of importance of predictors separately for the
two quintiles of interest for the purposes of this article. The importance ranking indicates which
predictors are important in forecasting a country being in the top or bottom quintile. The variable
importance for all of the quintiles is presented in Appendix 2. Only two variables have an
importance level over 50% for the top quintile: risk and electricity consumption (Table 5).
54 M. SKALETSKY ET AL.

Table 4. Descriptive Statistics of the Internet Use Rate for Quintiles.

Quintile Mean N Std. Deviation
Bottom 20% 3.55 34 2.21
2 11.99 33 2.74
3 30.57 34 6.15
4 51.56 33 8.98
Top 20% 79.76 33 7.46

Table 5. Variable Importance in Predicting Quintile Membership of Internet Penetration.

Variables/Quintile Relative Importance
Top quintile
Risk 100%
Electricity consumption per capita 59.38%
Bottom quintile
Number of telephone lines per 100 people 100%
GNI per capita 45.17%

Interestingly, the results for the bottom quintile are different. Only one variable has a level of
importance over 50%: the number of telephone lines per 100 people. The next most important
variable is GNI per capita, with an importance level of 45.17%.
The partial effect of each of the variables with a level of importance higher than 50% on the rate of
Internet use for countries in the top quintile (holding all other variables constant), is shown in Figure 4.
Figure 4 shows that the likelihood for a country to be placed in the top quintile increases sharply
after electricity consumption reaches approximately 5,000 Kwh per person. Top quintile membership
drops when the risk value reaches approximately 2.
The partial effect of the number of telephone lines per 100 people on the likelihood of being in the
bottom quintile is presented in Figure 5. The likelihood of being in the bottom quintile decreases
sharply when the number of telephone lines per 100 people reaches approximately 4 and when
income per capita reaches approximately 3,000. Once the number of telephone lines per 100 people
reaches approximately 10, it does not affect the likelihood of being in the bottom quintile.
Several interactions were found among predictors of the bottom Internet quintile membership
(Appendix 3). When electricity consumption levels in a country are low, risk has a sharper effect on
the likelihood of being in a bottom quintile. As the level of electricity consumption increases, the
effect of risk decreases.
The effect of primary school enrollment also depends on risk. When risk levels are over
approximately 2.5, primary school enrollment has no effect on the likelihood of being in the bottom
quintile until it reaches approximately 85%. For countries with risk value over three, the level of

Figure 4. Partial effect of each of the variables with level of importance higher than 50% on membership in the top quintile.
 JOURNAL OF GLOBAL INFORMATION TECHNOLOGY MANAGEMENT 55

Figure 5. Partial effect of each of the variables with the level of importance higher than 50% on membership in the bottom
quintile.

Table 6. Variable Importance for Mobile Phone Penetration.

Variables Relative Importance
GNI per capita 100%
Number of telephone lines per 100 people 82.06%
Electricity consumption per capita 65.51%

education does not have an effect until it reaches approximately 90%. Therefore, the education effect
on the likelihood of being in the bottom quintile increases when the level of infrastructure increases
and the level of risk decreases, which means that, to some extent high, risk effects can be alleviated
by better education and infrastructure.

Predictors of Mobile Phone Penetration

The next model explores the importance of the predictors of mobile phone use for all countries
included in the dataset. The three most important predictorsGNI per capita, the number of
telephone lines per 100 people and electricity consumptionare presented in Table 6. Figure 6
shows the partial effect of each of the predictors with the highest level of importance on mobile
phone use, after taking into consideration effects of all other variables.
Income affects the level of mobile phone use until it reaches approximately $20,000, at which
point the effect levels off. The number of telephone lines per 100 people affects the rate of mobile
phones until it reaches approximately 10 per 100 people. The level of mobile phone use increases
sharply with very minimal levels of electricity. After this sharp increase, electricity does not affect the
level of mobile phone use. These results indicate that even minimal levels of infrastructure result in
sharp increases in the level of mobile phone use.

Predictors of Mobile Quintile Membership

In the next model, the authors created a variable indicating to which quintile a country belongs, based
on the number of mobile phone users per 100 people. Quintile #5 includes countries with the highest
number of mobile phone users and quintile #1 includes countries with the lowest rate of mobile phone
users. Examples of countries in quintile #5 include Argentina, Singapore, and the United Kingdom. On
the other hand, similar to the bottom quintile for internet use, quintile #1 for mobile phone use has a
high proportion of African countries. Examples of countries in this group include Uganda,
56 M. SKALETSKY ET AL.

Figure 6. Partial effect of each of the variables with level of importance higher than 50% on the number of mobile phone
subscriptions per 100 people.

Table 7. Descriptive Statistics of the Number of Mobile Phone Subscriptions Per 100 People for Quintiles.
Quintile Mean N Std. Deviation
Bottom 20% 31.43 34 14.34
2 70.31 33 9.68
3 97.70 34 6.22
4 118.19 33 5.8
Top 20% 154.63 33 20.75

Mozambique, and Bangladesh. Also, similar to the Internet quintile model, quintile membership was
used as a dependent variable in this model. Descriptive statistics of the number of mobile phone users
per 100 people for different quintiles are shown in Table 7. The full list of countries included in the top
and the bottom quintiles is provided in Appendix 4. Interestingly, some countriessuch as Libya,
Argentina, and Kuwait, which have relatively low levels of Internet useare included in the top quintile,
based on the number of mobile phone subscriptions per 100 people. This can be attributed to a
substitution effect, where countries lacking the infrastructure necessary for the use of Internet and
landline telephones, substitute those technologies with mobile phones. Some countriessuch as Canada,
France, Japan, the United States, and some others, which have a relatively high level of Internet useare
not included in the top quintile, based on the number of mobile phone subscriptions per 100 people.
Variable importance for all of the quintiles is presented in Appendix 5. Interestingly, income per
capita is the most important predictor of membership in both top and bottom quintiles, as well as
that of the overall mobile phone use, as described in the previous section (Table 8). Also interest-
ingly, the three most important predictors of the number of mobile phone subscriptions are the same
as the most important predictors of membership in the bottom Internet quintile. The number of
telephone lines per 100 people, determined to be one of the most important predictors of bottom
 JOURNAL OF GLOBAL INFORMATION TECHNOLOGY MANAGEMENT 57

Table 8. Variable Importance in Predicting Quintile Membership of Number of Mobile Phone Subscriptions Per 100 People.
Variables/Quintile Relative Importance
Top quintile (quintile 5)
GNI per capita 100%
Electricity consumption per capita 69.22%
Number of telephone lines per 100 people 65.74%
Risk 61.89%
Secondary school enrollment 61.11%
Bottom quintile (quintile 1)
Electricity consumption per capita 100%
GNI per capita 99.69%
Number of telephone lines per 100 people 70.88%
Secondary school enrollment 59.61%

Table 9. Similarities and Differences With Prior Research.

Confirm Differ
Deichmann Consistent with the findings of Deichmann et al. (2006), While the Deichmann et al. study (2006) finds risk not
et al. the current study found income, education and to be an important predictor of Internet use, the
(2006) infrastructure to be important predictors of Internet use. current study found it to be almost as important as the
national income.
While Deichmann et al. find openness to trade to be an
important predictor of Internet use, the current study
found it to be unimportant.
Deichmann et al. find that the level of infrastructure
development has to be relatively high to lead to
increase in Internet use, while the current study found
that these levels can be relatively low to drastically
decrease the odds for a country to be in the bottom
Internet quintile.
Chinn and Find measure of regulatory quality to be an important Find negative effect of urbanization on the level of
Fairlie differential between levels of Internet penetration in the Internet penetration. The current study found this
(2007) United States and in the Middle Eastern region. The effect to be positive but not important.
authors found this measure to be the most important Find education not having a significant effect of
predictor of the level of Internet use overall and in differentiator between the United States and other
predicting a countrys placement into the top quintile regions. The authors found it to be a third important
based on the level of Internet use. factor for countries with the lowest levels of Internet
Find infrastructure to be an important differentiator use.
between the level of Internet use in the United States
and Sub-Saharan Africa. While the current study did not
differentiate countries based on geographical regions,
many African countries are included in the bottom
quintile in our study. The current study confirmed the
infrastructure to be an important predictor of a country
being placed in the bottom quintile based on the level
of Internet use.
Dewan Find the number of telephone lines to be a significant Find negative relationship between telephone lines
et al. predictor of the Internet use for countries in the bottom and Internet use for the top quintile. The current study
(2005) quintile. found that for countries in the top, as well as in the
bottom quintile, the more telephone lines, the higher
the likelihood for a country to be in the top quintile or
to move to a higher quintile.
Billon et al. Consistently with the findings of Billon et al. (2010), the The authors added to this study and build separate
(2010) current study found GDP to be important in predicting models for Internet and mobile phone penetration,
both Internet and mobile phones penetration. It was instead of using a digitalization index. The current
also found that the quality of regulation was the most study found that the quality of regulation is an
important predictor of Internet penetration for highly important predictor of Internet penetration, but is
digitalized countries. Infrastructure is found to be the much less important in predicting mobile phone
most important predictor of digital development for penetration.
countries with the lowest levels of digitalness. By employing TreeNet the authors were able to find
critical levels for each important predictor at which
levels of Internet and mobile phone use drastically
increase. It was also found that high risk effects can be
alleviated by better education and infrastructure.
58 M. SKALETSKY ET AL.

Figure 7. Partial effect of each of the variables with level of importance higher than 60% on membership in the top quintile of
number of mobile phone subscriptions.

quintile membership, is also one of the most important predictors of the number of mobile phone
subscriptions per 100 people overall. Besides income per capita, overall risk is found to be an
important predictor of top quintile membership and is also found to be the most important
predictor of membership in the top Internet quintile.
The partial effect of each of the variables with level of importance higher than 60% on member-
ship in the top quintile of the number of mobile subscriptions per 100 people, holding all other
variables constant, is shown in Figure 7.
Figure 7 shows that the likelihood for a country to be placed in the top quintile begins to increases
sharply as GNI increases until it reaches about $20,000 per capita; it levels off after that. The
likelihood of belonging to the top quintile increases sharply when the risk level reaches approxi-
mately 2, and drops when the risk value reaches approximately 2.5. This indicates that countries with
the lowest levels of risk are not necessarily those with the highest number of mobile phone
subscriptions. Interestingly, the likelihood of being in the top quintile, based on mobile phone use,
drops sharply when secondary school enrollment reaches approximately 80% and the number of
 JOURNAL OF GLOBAL INFORMATION TECHNOLOGY MANAGEMENT 59

Figure 8. Partial effect of each of the variables with level of importance higher than 60% on membership in the bottom quintile of
the number of mobile phone subscriptions.

telephone lines per 100 people reaches about 45. Thus, countries with the highest rates of mobile
phone use are not the countries with highest levels of infrastructure and education. This result can be
explained by the fact that this technology is appealing for many countries with relatively low levels of
institutional development, such as Saudi Arabia, Russia, and Kuwait.
Partial effects of each of the variables with level of importance higher than 50% for bottom
quintile membership are presented in Figure 8.
Figure 8 shows that the likelihood for a country to be in the bottom quintile decreases when income
per capita reaches approximately $2,000 and levels off at approximately $2,600. This indicates that
countries with higher income per capita mostly belong to quintiles with higher levels of mobile phone
use. A countrys likelihood of being in the bottom quintile decreases sharply once the number of
telephone lines per 100 people reaches approximately 2 and electricity consumption reaches approxi-
mately 1,000 Kwt per person. Thus, countries with the lowest levels of mobile phone use are those with
the lowest level of income per capita and infrastructure. The authors were unable to identify interactions
among the predictors of likelihood of being in the bottom or top quintile based on the level of mobile
phone use, as in the overall mobile phone use model.
Interestingly, very different results were found for the Internet penetration and Mobile phones
penetration models. It was also found that the factors that propel countries to the top and bottom of the
Internet and Mobile phone use are different. A detailed discussion of the results is given in the next section.

Discussion
Internet Penetration
TreeNet was employed to explore the nature of the relationships between Internet penetration and
its predictors for all of the countries. The importance of predictors which propel countries to top or
60 M. SKALETSKY ET AL.

bottom levels of digital development was also evaluated. Some differences and similarities with
previous studies were found, which is summarized in Table 9.
Interestingly, the current results indicate that the most important predictors of Internet
penetration are a country risk indicator and national income per capita. While national income
was found to be significant in many previous studies (Deichmann et al., 2006; Dewan et al., 2005;
Chinn & Fairlie, 2007), riskwhich is a comprehensive measure of a countrys institutional
development and stabilityhas been given much less attention previously. This is important to
policy makers in poorer developing countries: While national income presents a formidable
barrier, means by which institutional development and stability can be established may prove
to be a more feasible approach.
While institutional quality was long considered an important factor in national and regional
economic development (cf., Bevan et al., 2004; Chinn & Ito, 2006; Hasan et al., 2009), its
importance has been discussed to a lesser extent in DD research. Deichmann and colleagues
(2006) include a risk variable in their regression model but find it to be unimportant. Chinn and
Fairlie (2007) found a similar measure of regulatory quality to be important in differentiating
between the United States and Middle Eastern countries in the level of Internet penetration, but
not to be important overall. Billon and colleagues (2010) find high regulation quality to have a
positive influence on the digitalization index in countries with high levels of digital development.
Recently, institutional quality has received more attention in the context of e-commerce.
Martinez and colleagues (2010), for example, argue that national institutional quality is a more
important antecedent of e-commerce development (which depends on the level of Internet
penetration), than the level of entrepreneurial activity. The current finding that institutional
quality is as important as the level of national income in predicting the level of Internet use
supports a stronger emphasis on institutional development for Internet penetration and
e-commerce.
However, it was also found that, while institutional development is one of the most important
predictors of the level of the Internet use overall for all of the countries and the most important
predictor of a country being placed in the top quintile based on the level of Internet use, it is not an
important predictor of a country being placed in the bottom quintile. In fact, it is less important than
all of the measures of infrastructure, demographic characteristics, education, and income. While a
countrys placement into the top quintile, based on Internet use, is determined mostly by the level of
a countrys risk, placement in the bottom quintile, based on the level of Internet use, is determined
mostly by the level of infrastructural development. Thus, it was concluded that infrastructure is
important for initial digital development. However, a country cannot move into a higher quintile
unless it develops the quality of its institutions.
While the importance of infrastructure development has been discussed in past literature, the
authors were able to clearly demonstrate the differences in the importance of infrastructure,
country risk and other measures for a countrys placement in a top or a bottom quintile of
Internet use. Critical levels of infrastructure, education and other predictors at which level of
Internet use seems to drastically increase were also able to be identified. The current results
demonstrate that even basic levels of infrastructure developmentsuch as four landline telephones
per 100 peopledecreases the likelihood of a country being placed in the bottom quintile. While
this does not establish a cause and effect relationship, it can still be informative for policymakers
when making decisions regarding investment into infrastructure development in countries with
the lowest levels of Internet use. It was found that the effect of some of the predictors on
membership in the bottom Internet use quintilesuch as educationdepends on levels of infra-
structure and the level of country risk. The education effect on the likelihood of being in the
bottom quintile increases when the level of infrastructure increases and the level of risk decreases.
This means that high risk effects can be alleviated by better education and infrastructure. These
findings provide important indications as to which factors may help poorer countries achieve
higher levels of digital development.
 JOURNAL OF GLOBAL INFORMATION TECHNOLOGY MANAGEMENT 61

Mobile Phone Penetration

Separate models were built predicting the level of countries mobile phone use and likelihood of being
in the top and bottom quintiles based on mobile phone use. This technology was not included in
studies by Deichmann et al. (2006) and Dewan et al. (2005). We find that mobile phone penetration is
predicted by very different factors from those predicting Internet penetration. In contrast to the
Internet penetration model, the most important overall predictors of the level of mobile phone use
and predictors of a countrys placement in the top and the bottom quintiles are the same, based on the
number of mobile phone subscriptions per 100 people. The three most important predictors are
income per capita, number of telephone lines per 100 people and level of electricity consumption.
It was found that most of the countries placed in the bottom quintile by the level of Internet use
are also in the bottom quintile based on the number of mobile phone subscriptions. Countries in the
top quintile, however, are very different; only eight countries are common to both top quintiles. The
profile of these countries is also very different. Countries with the highest number of mobile phone
users have lower levels of infrastructure and income than countries with the highest levels of Internet
use, but they have relatively high levels of education and proportion of urban population. These
countries also have a relatively high level of risk, indicating a lack of institutional development.
Average values of all of the predictors for counties in the top and bottom quintiles, based on the
number of mobile phone subscriptions and on the number of Internet users per 100 people, are
shown in Appendix 6.
These findings confirm that, overall, mobile phone penetration is relatively independent of other
technologies, in particular Internet penetration. Lower cost and less expensive infrastructure needed
for mobile phones make this technology appealing and affordable for many countries that do not
have widespread Internet use and high levels of income per capita. Mobile phone use has increased
drastically in less economically developed countries, particularly in Africa, in recent years (Hosman
& Fife, 2012; Soremekun & Malgwi, 2012). However, as summary statistics in Table 7 show, many
countries still lag significantly behind in the level of mobile phone use. It was found that minimal
levels of infrastructure are sufficient for a sharp increase in mobile phone use, after which infra-
structure does not have an effect on the rate of mobile phone subscriptions. The authors were not
able to identify interactions among the predictors, confirming that mobile technology is less
dependent on factors that were found important for Internet penetration. While the use of a mobile
phone itself does not substitute need for the Internet, as smart phones become more common and
less expensive, the use of Internet in countries with high levels of mobile phone use will increase.
Thus, it was concluded that mobile phones provide an important platform for increasing the levels of
Internet use and closing the DD for countries with medium levels of economic development.
However, it was found that countries with the lowest levels of Internet use also have the lowest
levels of mobile phone use. These countries lack the basic infrastructure necessary to support the use
of ICT. These are the countries that require the attention of policy makers. The finding that critical
levels of infrastructure, after which the level of mobile phone use drastically increases and after
which the likelihood of being in the bottom quintile sharply decreases, are very low. Thus, it was
concluded that policies should aim at building basic infrastructure, which would jumpstart the use of
ICT. The authors find that critical levels of infrastructure that lead to sharp increases in the use of
mobile phones are lower than critical levels at which Internet use sharply increases. Thus, policies
aimed at the promotion of ICT in the least developed countries should concentrate on building the
basic infrastructure needed to support the use of mobile phones, which in turn would lead to an
increase of Internet use via smart phone technologies.

Limitations and Future Perspectives

This study has a number of limitations. The current study was limited by data availability in both the
number of countries and in types of measures available. The present analysis is limited to quantitative
62 M. SKALETSKY ET AL.

measures such as the number of Internet users per 100 people but not the quality of the Internet
connection or the ways in which it is used. National level data with country as a unit of analysis was
used, which may present some accuracy and comparability issues (Kotabe & Cunningham, 2002). By
doing so, complexities and diversities within countries are unable to be accounted for.
There are several possibilities for future research. First, the authors would like to expand the
dataset to include multiple years. A longitudinal study could provide additional insights on how the
importance of predictors of the global DD changes over time. Another possibility for future research
is to analyze penetration of modern technologies, such as cloud computing.

Conclusion
This study contributes to the understanding of the importance and complexity of the predictors of the
global DD, which answers calls by Desouza et al. (2006, 2007) for IS research to consider key societal
questions. While the DD has been addressed in many past studies (e.g., Chinn & Fairlie, 2007; Deichmann
et al., 2006; Dewan et al., 2005), the use of TreeNet allowed for obtaining a greater level of detail with
regard to these relationships. Critical levels of infrastructure at which a country is likely to move to a
higher quintile was identified based on the level of Internet and mobile phone use (e.g., the likelihood of
being in the bottom quintile decreases sharply when the number of telephone lines per 100 people reaches
approximately 4). It was discovered that very minimal levels of infrastructure are needed in order to
jumpstart the use of mobile phones (e.g., the minimal number of telephone lines per 100 people affects the
rate of mobile phones until it reaches approximately 10 per 100 people, and very minimal levels of
electricity consumption sharply increase the level of mobile phone use). These are new findings that have
not been addressed by past studies but can inform policy decisions regarding investments in infrastruc-
ture development of countries where this is most needed. It was demonstrated that required critical levels
are lower for the increase in the use of cell phones than for the increase in Internet use. These findings
provide information to policymakers regarding the specific levels of basic infrastructure needed to
jumpstart the use of Internet and mobile phones in the least developed countries.

Note
1. IHSInformation Handling Services (www.ihsglobalinsight.com).

References
Andonova, V. (2006). Mobile phones, the Internet, and the institutional environment. Telecommunications Policy, 30
(1), 2945. doi:10.1016/j.telpol.2005.06.015
Bagchi, K. (2005). Factors contributing to global digital divide: Some empirical results. Journal of Global Information
Technology Management, 8(3), 4765. doi:10.1080/1097198X.2005.10856402
Bagchi, K., Hart, P., & Peterson, M. F. (2004). National culture and information technology product adoption. Journal
of Global Information Technology Management, 7(4), 2946. doi:10.1080/1097198X.2004.10856383
Bagchi, K., & Udo, G. (2007). Empirically testing factors that drive ICT adoption in Africa and OECD set of nations.
Issues in Information Systems, 8(2), 4552.
Bankole, F. O., Osei-Bryson, K., & Brown, I. (2013). The impact of ICT investments on human development: A
regression splines analysis. Journal of Global Information Technology Management, 16(2), 5985. doi:10.1080/
1097198X.2013.10845636
Beilock, R., & Dimitrova, D. V. (2003). An exploratory model of inter-country Internet diffusion. Telecommunications
Policy, 27(34), 237252. doi:10.1016/S0308-5961(02)00100-3
Bevan, A., Estrin, S., & Meyer, K. (2004). Foreign investment location and institutional development in transition
economies. International Business Review, 13(1), 4364. doi:10.1016/j.ibusrev.2003.05.005
Billon, M., Lera-Lopez, F., & Marco, R. (2010). Differences in digitalization levels: A multivariate analysis studying the
global digital divide. Review of World Economics, 146(1), 3973. doi:10.1007/s10290-009-0045-y
Billon, M., Marco, R., & Lera-Lopez, F. (2009). Disparities in ICT adoption: A multidimensional approach to study the
cross-country digital divide. Telecommunications Policy, 33(1011), 596610. doi:10.1016/j.telpol.2009.08.006
Chen, W., & Wellman, B. (2004). The global digital dividewithin and between countries. IT & Society, 1(7), 3945.
 JOURNAL OF GLOBAL INFORMATION TECHNOLOGY MANAGEMENT 63

Chinn, M. D., & Fairlie, R. W. (2007). The determinants of the global digital divide: A cross-country analysis of
computer and internet penetration. Oxford Economic Papers-New Series, 59(1), 1644. doi:10.1093/oep/gpl024
Chinn, M. D., & Ito, H. (2006). What matters for financial development? Capital controls, institutions, and interac-
tions. Journal of Development Economics, 81(1), 163192. doi:10.1016/j.jdeveco.2005.05.010
Dasgupta, S., Lall, S., & Wheeler, D. (2005). Policy reform, economic growth and the digital divide. Oxford
Development Studies, 33(2), 229243.
Deichmann, J., Eshghi, A., Haughton, D., Masnaghetti, M., Sayek, S., & Topi, H. (2006). Exploring break-points and
interaction effects among predictors of the international digital divide: An analytical approach. Journal of Global
Information Technology Management, 9(4), 4771. doi:10.1080/1097198X.2006.10856432
Desouza, K. C., Ein-Dor, P., McCubbrey, D. J., Galliers, R. D., Myers, M. D., & Watson, R. T. (2007). Social activism in
information systems research: Making the world a better place. Communications of the Association for Information
Systems, 19, 261277.
Desouza, K. C., El-Sawy, O. A., Galliers, R. D., Loebbecke, C., & Watson, R. T. (2006). Beyond rigor and relevance
towards responsibility and reverberation: Information systems research that really matters. Communications of the
Association for Information Systems, 17(16), 341353.
Dewan, S., Ganley, D., & Kraemer, K. L. (2005). Across the digital divide: A cross-country multi-technology analysis of
the determinants of IT penetration. Journal of the Association for Information Systems, 6(12), 409432.
Dismuke, C., & Lindrooth, R. (2006). Ordinary least squares. Methods and Designs for Outcomes Research, 93, 93104.
Eshghi, A., Haughton, D., Li, M., Senna, L., Skaletsky, M., & Woolford, S. (2011). Enrollment management in graduate
business programs: Predicting student retention. Journal of Institutional Research, 16, 2.
Friedman, J. (1991). Multivariate adaptive regression splines. The Annals of Statistics, 19(1), 167. doi:10.1214/aos/
1176347963
Friedman, J. H. (2001). Greedy function approximation: A gradient boosting machine. Annals of Statistics, 29, 1189
1232.
Hargittai, E. (1999). Weaving the Western Web: Explaining differences in internet connectivity among OECD
countries. Telecommunications Policy, 23(1011), 701718. doi:10.1016/S0308-5961(99)00050-6
Hasan, I., Wachtel, P., & Zhou, M. (2009). Institutional development, financial deepening, and economic growth:
Evidence from China. Journal of Banking & Finance, 33(1), 157170. doi:10.1016/j.jbankfin.2007.11.016
Hosman, L., & Fife, E. (2012). The potential and limits of mobile phone usage for development in Africa: Innovation
and top-down-meets-bottom-up partnering. The Journal of Community Informatics, 8, 3.
International Telecommunication Union. (2009). Measuring the information society. Geneva, Switzerland: Author.
Available at http://www.itu.int/en/ITU-D/Statistics/Documents/publications/mis2014/MIS2014_without_Annex_4.
pdf
Jann, B. (2008). The Blinder-Oaxaca decomposition for linear regression models. The Stata Journal, 8(4), 453479.
Kiiski, S., & Pohjola, M. (2002). Cross-country diffusion of the Internet. Information Economics and Policy, 14(2), 297
310. doi:10.1016/S0167-6245(01)00071-3
Koenker, R., & Hallock, K. (2001). Quantile regression: An introduction. Journal of Economic Perspectives, 15(4), 143
156. doi:10.1257/jep.15.4.143
Kotabe, M., & Cunningham, P. (2002). Using Euromonitor database in international marketing research. Journal of the
Academy of Marketing Science, 30(2), 172.
Lerman, R. I., & Yitzhaki, S. (1984). A note on the calculation and interpretation of the gini index. Economics Letters,
15(34), 363368. doi:10.1016/0165-1765(84)90126-5
Lu, M.-T. (2001). Digital divide in developing countries. Journal of Global Information Technology Management, 4(3),
14. doi:10.1080/1097198X.2001.10856304
Martinez, C. A., & Williams, C. (2010). National institutions, entrepreneurship, and global ICT adoption: A cross-
country test of computering theories. Journal of Electronic Commerce Research, 11(1), 7391.
Norris, P. (2001). Digital divide? Civic engagement, information poverty, and the internet in the democratic societies.
New York: Cambridge University Press.
Pick, J., & Azari, R. (2011). A global model of technological utilization based on governmental, business-investment,
social, and economic factors. Journal of Management Information Systems, 28(1), 4984. doi:10.2753/MIS0742-
1222280103
Sciadas, G. (2005). From the digital divide to digital opportunities. Quebec, Canada: Claude-Yves Charron.
Soremekun, O., & Malgwi, C. A. (2012). Exploring the relationship between foreign direct investment and mobile
technology in Africa: An application of directed acyclic graphs. Case Studies In Business, Industry, and Government
Statistics, 5(1), 5866.
Steinmueller, E. (2001). ICTs and the possibilities for leapfrogging by developing countries. International Labour
Review, 140(2), 193210. doi:10.1111/ilr.2001.140.issue-2
World Bank. (2011). World Development Indicators 2011. Washington, DC: Author.
The World Factbook. (n.d.). Available at https://www.cia.gov/library/publications/the-world-factbook/. Washington,
DC: Central Intelligence Agency.
64 M. SKALETSKY ET AL.

Author Notes
Maria Skaletsky is a research consultant at Bentley University. She recently earned her PhD in business analytics from
Bentley, where her research focused on living standards and the digital divide. She also holds an MBA (concentration
in business data analysis) from Bentley University and a BS in business management from St. Petersburg Academy of
Business (Russia). Fluent in Russian as well as English, she has published on how the digital divide hasand has not
changed in different parts of the world in recent years in Information Technology for Development and has also studied
the phenomenon within the Russian Federation.

Robert D. Galliers became Bentley Universitys inaugural University Distinguished Professor in 2009, having served as
Provost since 2002 in which role he headed up Bentleys progress to university status and EQUIS and EDAMBA
accreditations. He also holds a fractional appointment as professor of information systems in the School of Business &
Economics at Loughborough University, United Kingdom. He received the AISs LEO Award for exceptional lifetime
achievement in IS in 2012 and was awarded an Honorary Doctor of Science degree by Turku University, Finland in
1995. He has been keynote speaker at more than 80 major international conferences and symposia including the
Australian, European, Mediterranean, Scandinavian, and UKAIS Conferences on IS. He has over 300 publications to
his name, including over 90 journal articles and 12 booksthe most recent of which being Critical Perspectives on
Business and Management: Management Information Systems (Taylor & Francis, 2015) and The Oxford Handbook of
Management Information Systems (Oxford University Press, 2011). His work has been cited over 8,000 times and is
trans-disciplinary in nature, focusing primarily on organizational innovation/transformation; the processes and
practices of IS strategizing; knowing in organizations, and the intra- and extra-organizational impacts of ICT. He is
the founding editor-in-chief of The Journal of Strategic Information Systemsone of eight journals in the AIS Senior
Scholars basket of leading IS journals.

Dominique Haughton (Ph.D., MIT 1983) is professor of mathematical sciences and global studies and graduate
coordinator for business analytics at Bentley University, as well as an affiliated researcher at the Universit Paris 1 and
Universit Toulouse 1, France. Her areas of interest are business analytics, living standards analytics, and data mining.
She is a United States co-editor of Case Studies in Business, Industry, and Government Statistics (CSBIGS). She is also
the author of over 60 articles and a book on living standards analytics. Dr. Haughton is also a Fellow of the American
Statistical Association.

Olumayokun Soremekun is the head of business development at OLX in Lagos, Nigeria. She holds a PhD from Bentley
University, a Master of Science from Western Carolina University, and a Bachelor of Arts from Tusculum College. She
has worked on consulting projects with the World Bank and the United Nations.

Appendix 1. List of Countries for the Top and Bottom Quintiles Based on the Rate of Internet Use.
Quintile Countries
Top Austria, Antigua and Barbuda, Australia, Barbados, Belgium, Bermuda, Canada, Denmark, Estonia, Finland, France,
Germany, Hong Kong, Iceland, Ireland, Japan, Korea, Latvia, Luxemburg, Norway, Netherlands, New Zealand, Qatar,
Sweden, Switzerland, Slovak Republic, St. Kitts and Nevis, Singapore, St. Vincent and the Grenadines, Slovenia, United
Kingdom, United States, UAE.
Bottom Bangladesh, Benin, Botswana, Burkina Faso, Cambodia, Congo (Rep), Congo (Dem. Rep), Cameroon, Cote dIvoire,
Chad, Central African Republic, Comoros, Djibouti, Eritrea, Ethiopia, Gabon, Guinea-Bissau, Guinea, India, Liberia, Lao
PDR, Lesotho, Mozambique, Mauritania, Myanmar, Mali, Malawi, Madagascar, Namibia, Niger, Papua New Guinea,
Solomon Islands, Turkmenistan, Togo.

Appendix 2. Variable Importance in Predicting Countries Internet Quintile Membership.

Variable Importance
Variable Name Quintile 1 (bottom) Quintile 2 Quintile 3 Quintile 4 Quintile 5 (top)
GNI per capita 45.17% 95.45% 87.03% 68.18% 54.09%
Proportion of Urban population 25.69% 60.32% 46.94% 37.68% 13.05%
Secondary school enrollment 41.45% 56.68% 100% 52.06% 14.24%
Current Overall Risk 25.19% 100% 96.87% 100% 100%
Primary school enrollment 44.21% 75.97% 40.69% 46.53% 11.64%
Electricity consumption per capita 41.08% 68.56% 81.52% 63.55% 59.38%
GINI Index 22.84% 57.87% 79.56% 44.40% 16.52%
Price of a 3 min phone call 24.26% 61.26% 74.25% 34.62% 10.79%
Proportion of population ages 1564 28.07% 64.03% 59.81% 85.12% 12.43%
Proportion of population ages 65 and above 31.90% 89.47% 78.36% 58.44% 21.98%
Merchandise trade as % of GDP 22.72% 55.31% 56.67% 38.19% 10.91%
Number of telephone lines per 100 people 100% 79.01% 74.39% 63.79% 51.58%
 JOURNAL OF GLOBAL INFORMATION TECHNOLOGY MANAGEMENT 65

Appendix 3
Interactions of Predictors of a Country Membership in a Bottom Quintile Based on the Level
of Internet Use Electricity Consumption per Capita and Risk
66 M. SKALETSKY ET AL.

Appendix 3 (Continued)
 JOURNAL OF GLOBAL INFORMATION TECHNOLOGY MANAGEMENT 67

Appendix 4. List of Countries for the Top and Bottom Quintiles Based on the Rate of Mobile Phone Use.
Quintile Countries
Top Albania, Antigua and Barbuda, Argentina, Austria, Bermuda, Bulgaria, Croatia, Czech Republic, Finland, Hong Kong,
Israel, Italy, Kuwait, Libya, Lithuania, Luxemburg, Macao SAR, Maldives, Oman, Panama, Portugal, Qatar, Russia, Saudi
Arabia, Seychelles, Singapore, St. Kitts and Nevis, Suriname, Trinidad and Tobago, UAE, United Kingdom, Uruguay,
Vietnam.
Bottom Zambia, Yemen, Uganda, Tonga, Togo, Tanzania, Sudan, Solomon Islands, Rwanda, Papua New Guinea, Niger, Nepal,
Myanmar, Mozambique, Mali, Malawi, Madagascar, Liberia, Lesotho, Haiti, Guinea-Bissau, Guinea, Ethiopia, Eritrea,
Djibouti, Cuba, Congo (Dem. Rep), Comoros, Chad, Central African Republic, Cameroon, Burkina Faso, Bangladesh,
Angola.

Appendix 5. Variable Importance in Predicting Countries Mobile Phone Subscriptions Quintile Membership.
Variable Importance
Variable Name Quintile 1 (Bottom) Quintile 2 Quintile 3 Quintile 4 Quintile 5 (Top)
GNI per capita 99.69% 100% 69.04% 66% 100%
Proportion of urban population 42.66% 71.24% 73.17% 61.53% 47.22%
Secondary school enrollment 59.61% 47.26% 51.31% 100% 61.11%
Current overall risk 51.25% 67.44% 42.64% 89.73% 61.89%
Primary school enrollment 17.24% 43.86% 74.83% 58.43% 60.26%
Electricity consumption per capita 100% 67.10% 95.42% 66.74% 69.22%
GINI Index 16.00% 79.02% 68.32% 93.19% 56.26%
Price of a 3 min. phone call 13.14% 53.85% 53.84% 55.62% 29.77%
Proportion of population ages 1564 33.88% 51.85% 65.32% 49.95% 42.23%
Proportion of population ages 65 and above 31.86% 76.18% 81.18% 67.20% 43.81%
Merchandise trade as % of GDP 18.26% 60.72% 100% 77.32% 39.58%
Number of telephone lines per 100 people 70.88% 74.26% 62.69% 60.34% 65.74%

Appendix 6. Average Values of the Predictors for Countries in Different Quintiles.

Internet Users Quintiles Mobile Phones Subscriptions Quintiles
Top Bottom Top Bottom
Telephone lines per 100 people 43.46 1.88 30.85 2.31
GINI index 32.80 43.74 38.82 42.97
GNI per capita 35827.51 2555 27320.24 1683.74
Electric power consumption kWh per capita 10046.44 324.81 6104.6 180.05
Merchandise trade of GDP 89.74 65.69 87.17 58.97
Population ages 1564 68.6 57.04 69.17 55.98
Population ages 65 and above 14.07 3.4 10.75 3.55
Urban population 74.79 37.68 71.44 34.21
School enrollment primary 94.87 75.29 92.47 75.40
School enrollment secondary 86.91 33.36 80.30 31.63
Current overall risk 1.78 3.56 2.28 3.71
Percent of households with computer 78.32 3.93 58.26 3.54
Price of a 3 min. local call 0.14 0.18 0.11 0.18