WEEK 6/P/ 2 BEHS 103
“Bad Neighborhood” and Internet Adoption in Poor Countries: What is behind the Persistent Digital Gap?
DIRK DOHSE AND CHENG YEE LIM
ABSTRACT The paper investigates the determinants of Internet adoption in poor countries, focusing on the role of
macro-geographic location (neighborhood). It is argued that neighboring countries are interconnected by various kinds of
spillovers, including knowledge spillovers as well as spillovers of norms and attitudes that affect individual adoption
behavior. The empirical findings support the view that Internet adoption is affected by adoption rates in neighboring coun-
tries, even when controlling for a wide range of covariates. Addressing potential endogeneity concerns using an instru-
mental variable approach moreover suggests these relationships to be causal. The findings imply that international policies
to support Internet adoption in poor countries might be more effective if they target groups of neighboring countries rather
than single countries in order to better exploit spillovers between neighboring countries.
Motivation
D igitization is rewriting the rules of international competition, bringing about manifold opportu-nities for newcomers to enter global value chains and to catch up with incumbents. This applies not only to companies, but, in principle, also to countries. It is, however, by no means clear
whether the digital revolution will help poor countries to better integrate into the global economy
and to catch up in terms of income and wealth, or whether the rich countries will be able to sustain
or even accelerate their competitive advantage by means of digital technologies (World Bank 2016a).
The core enabling technology indispensable for reaping the fruits of the digital revolution is the
Internet. Digital capabilities and, in particular, the capability to productively use the Internet increas-
ingly determine which companies, industries, and countries create or lose value (Capello and
Nijkamp 1996a,b; Hirt and Willmot 2014). It is, therefore, of critical importance that developing
countries swiftly abridge the digital gap that separates them from developed economies. Empirical
reality looks different, however. As can be seen from Figure 1 and Table 1, the differential in Internet
usage between developed and developing countries has in fact widened from 43.1 to 47.4 percent
between 2005 and 2013.1 The rise of the digital gap2 is even more pronounced if one compares the
developed countries with Africa, the poorest continent on earth (Table 1). Hence, contrary to the rosy
picture of the Internet enabling new possibilities in communication and productivity in developing
countries, the benefits from information technologies may be widening the chasm between richer
nations and those that lack the infrastructure, skills, and resources (Norris 2001; Warf 2001). “Access
to the Internet is deeply conditioned by where one is” (Warf 2001: 16), and there is little indication
that the importance of geography is decreasing in the digital age. The point of the ICT-inequality
Dirk Dohse is a Senior Researcher in the Kiel Institute for the World Economy, Kiellinie 66, Kiel 24105, Germany.
His e-mail address is: dirk.dohse@ifw-kiel.de. Cheng Yee Lim is a graduate student in the University of Chicago, 5801
South Ellis Avenue, Chicago, IL 60637, USA. Her e-mail address is: limchengyee@outlook.com. An earlier version of this
paper (Dohse and Lim 2016) is available as a working paper. The authors are grateful to the editor and two anonymous
referees for most helpful comments. The usual disclaimer applies.
Submitted December 2016; revised May 2017; accepted July 2017.
VC 2017 Wiley Periodicals, Inc
Growth and Change DOI: 10.1111/grow.12220 Vol. 49 No. 1 (March 2018), pp. 241–262
nexus is well put by Rodriguez and Wilson, who argue that “when a new technology is introduced
into a social setting where scarce resources and opportunities are distributed asymmetrically, the
greater likelihood is that those with more resources will employ them to gain additional one, includ-
ing ICTs” (Rodriguez and Wilson 2000: 11).
Moreover, access to the Internet is more multidimensional than adoption of telephones, televi-
sions, and radios in the past. The binary division of users and non-users only captures one facet of
Internet access. DiMaggio et al. (2001) define the digital divide as “inequalities in access to the Inter-
net, extent of use, knowledge of search strategies, quality of technical connections and social support,
ability to evaluate the quality of information, and diversity of uses” (DiMaggio et al. 2001: 310).
Thus, the quality of use of the Internet will also result in a second level digital divide among users
(Hargittai 1999), reiterating concerns that the lack of digitalization may further marginalize develop-
ing countries from mainstream economic growth (Davison et al. 2000).
Although there is a rich and growing literature dealing with the determinants of Internet adoption
and the digital divide (discussed in more detail in the second section), the possibility of cross-country
interactions in the adoption process has attracted relatively little attention so far. The current paper
contributes to a better understanding of the effects of macro-geographic location (and of urban struc-
ture within countries) on Internet adoption in poor countries. Our core hypothesis is that, apart from
the usual suspects (including per capita income, education, telecommunications infrastructure, institu-
tions), the neighborhood of a country, i.e., its geographic location, has a crucial impact on the pro-
pensity of the country’s population to adopt and productively use the Internet.
The paper is structured as follows: We begin with a brief review of the pertinent literature in the
second section and develop our basic hypotheses in the third section. The fourth section presents the
FIGURE 1. INTERNET ADOPTION RATES IN DEVELOPED AND DEVELOPING COUNTRIES 2005–2013.
Source: ITU (2016), own compilation.
TABLE 1. DIGITAL GAP (IN PERCENTAGE POINTS) BETWEEN DEVELOPED AND DEVELOPING COUNTRIES.
Year 2005 2006 2007 2008 2009 2010 2011 2012 2013
Gap Developed–Developing 43.1 44.1 47.1 46.7 45.5 45.4 43.6 46.8 47.4
Gap Developed–Africa 47.4 49.0 53.9 55.0 56.3 57.0 56.4 61.1 62.5
Source: ITU (2016), own compilation.
242 GROWTH AND CHANGE, MARCH 2018
data and introduces our empirical strategy. The fifth section presents and discusses the empirical
results. The sixth section concludes and provides an outlook for policy and for future research.
Literature Review The existing literature on Internet diffusion has established three main channels—economic,
demographic, and institutional—to attain widespread Internet diffusion. Unsurprisingly, economic
variables, such as GDP per capita and telecommunication infrastructure, have been found to be key
determinants of Internet diffusion in numerous studies (see, e.g., Andres et al. 2010; Beilock and
Dimitrova 2003; Crenshaw and Robison 2006 Warf 2009; Wunnava and Leiter 2009). Hargittai
(1999) is one of the first econometric studies analyzing the spread of the Internet across countries.
She uses a sample of OECD countries for the period 1994–1997 and finds that even among OECD
countries that have similar levels of social and economic development, differences in GDP per capita
matter when it comes to predicting diffusion rates. Other important determinants of Internet diffusion
identified by Hargittai include telephone density and telecommunications policy. Kiiski and Pohjola
(2002) examined the determinants of Internet hosts per 1,000 inhabitants in 60 OCED and develop-
ing countries for the years 1995–2000 and found income per capita, telephone access costs and years
of schooling to be significant determinants. Beilock and Dimitrova (2003) extend the focus of
research on a large sample of 105 countries,3 including economies with vastly different socioeco-
nomic levels of development. Their results confirm that income per capita is, by far, the most impor-
tant determinant of Internet usage rates. Other factors found to be important include telephone
infrastructure and the political and economic openness of a country. Chinn and Fairlie (2007) ana-
lyzed the determinants of Internet penetration in a large cross section of countries and tried to decom-
pose their relative importance. They examined a panel of 161 countries over the 1999–2001 period
and found that the largest contributing factor to the Internet diffusion is per capita income, followed
by per capita telephone lines. In a follow-up study, focusing on differences between developing and
developed countries, Chinn and Fairlie (2010) found that the main factors responsible for low Inter-
net penetration rates in developing countries are disparities in per capita income, telephone density,
legal quality, and human capital.4
Other economic variables that have been investigated in different comparative studies include edu-
cational attainment, trade openness, and telecommunication regulatory policies. Mixed conclusions
have been found for these three variables, depending on the data used to measure them and on the
empirical specification. Crenshaw and Robison (2006) investigated the main factors contributing to
the change in the number of Internet hosts for 80 countries in the 1995–2000 period. One of their
main findings is that a country’s openness and its participation in global networks have a positive
impact on Internet development, whereas countries that remain “isolates” in the global system are at
a large disadvantage. Low levels or the lack of education are expected to hinder the accessibility and
diffusion of the technology. In Kiiski and Pohjola’s (2002) study of Internet diffusion in developing
and OECD countries, average years of schooling significantly affected Internet diffusion positively,
but telecom competition was found to be insignificant. Caselli and Coleman’s (2001) study of com-
puter diffusion found evidence that the attainment of secondary education and imports per worker
strongly increases computer diffusion. More recent work by Cruz-Jesus, Oliveira, and Bacao (2012),
Cruz-Jesus et al. (2016), and Tengtrakul and Peha (2013) confirms the importance of school educa-
tion for ICT adoption. Conversely, Chinn and Fairlie (2007) found no significant relationship
between education and Internet diffusion, but highlighted the importance of telecommunication regu-
latory policy as a determinant of Internet diffusion.
“BAD NEIGHBORHOOD” AND INTERNET ADOPTION 243
Besides economic factors, a number of studies also investigated the importance of institutions in
crafting and enforcing policies in ICTs advancement. Wallsten (2005) used a 45-country data set
from 2001 and found that stricter regulation of Internet service providers has a negative impact on
Internet usage. Andonova (2006) analyzed a cross-section of developed and developing countries for
a single year (2001). She used three different measures of institutional quality, namely political
rights, civil liberties, and political constraints, and their resultant effects on investment climate to
explain differences in mobile phone and Internet usage, and found a positive relationship between
institutional environment, infrastructural development, and Internet usage. Choucri, Madnick, and
Ferwerda (2014) identified cybersecurity as an increasingly important institutional parameter for ICT
development. Furthermore, countries with higher political freedom and better human and property
rights protection tend to have higher Internet adoption rates (Crenshaw and Robison 2006; Warf
2009).
Demographic controls were also included in several studies of Internet diffusion since certain
demographic characteristics are expected to push Internet adoption. Countries with greater urbaniza-
tion and a younger population are expected to adopt the Internet more readily. Studies by Chinn and
Fairlie (2007), Goldfarb and Prince (2008), and Niehaves and Plattfaut (2014) hint at a positive
impact of youth on Internet adoption, which is in line with findings from microdata (US Department
of Commerce 2002). Results with respect to urbanization are less clear-cut, however. Chinn and
Fairlie (2007), Howard and Mazaheri (2009), and Chinn and Fairlie (2010) find a significantly nega-
tive impact of urbanization (measured as the ratio of urban population in the total population),
whereas others (e.g., Andonova 2006; Crenshaw and Robison 2006) find opposite results. We argue
that using a single urbanization variable—as is usual in the literature—might be inadequate and pro-
pose an additional measure to better reflect urban structure in the next section.
While the majority of earlier work on Internet adoption is focused on developed and emerging
economies, there is currently an increasing interest in Internet adoption in poor and remote countries
(e.g., Oyeleye, Sanni, and Shittu 2015; P�enard et al. 2015; Priluck 2016; Touray, Salminen, and
Mursu 2015; West 2015). Most of these recent papers are, however, focused on single countries,
and there is relatively little systematic cross-country evidence for large samples of poor (i.e., low and
lower middle income) countries.5 The current paper contributes to fill this gap in the literature, mak-
ing use of a rich panel data set for 63 countries and a 9-year period (2005–2013), and focusing on
the impact of macro-geographic location (neighborhood).
The Spatial Dimension: Macro-Geographic Location and Urbanization Although there is an extensive literature examining cross-country Internet diffusion, the role of
macro-geographic location and the impact of neighboring countries on Internet adoption have not
been well explored as yet. This neglect is surprising as the importance of macro-geographic neigh-
borhood effects is well established in the knowledge spillover literature (e.g., Keller 2002) as well as
in development economics (e.g., Collier 2007). Moreover, the role of urbanization in Internet adop-
tion in poor countries is still unsettled. The current paper addresses these apparent gaps in the litera-
ture. This section puts forward the theoretical argument, which is exposed to rigorous econometric
analysis in the subsequent sections.6
Macro-geographic location (Neighborhood). The majority of empirical work on Internet
adoption has treated adoption units (countries) as independent and “ignored the possibility of cross-
country interactions in the adoption process” (Comin and Mestieri 2013: 29). There are, however,
good reasons to assume that Internet adoption in a country is affected by Internet adoption in
244 GROWTH AND CHANGE, MARCH 2018
neighboring countries. As is well-established in the pertinent literature, spatial proximity facilitates
the flow of knowledge. Spatial proximity (neighborhood) increases the likelihood of an encounter
between inhabitants of country A and country B. The higher the share of Internet users in country A,
the higher the likelihood that such an encounter will lead an inhabitant of neighboring country B to
adopt the new technology. As Keller (2002) has shown, international technology diffusion is geo-
graphically localized, implying that neighboring countries tend to benefit more (in terms of produc-
tivity gains) from innovation in a given country than more distant countries.7
Knowledge spillovers might, however, be not the only relevant kind of spillovers between neigh-
boring countries. According to the theory of planned behavior (Ajzen 1991), which has been widely
and successfully applied to predict behavioral intention in technology acceptance,8 an individual’s
behavioral intention in a specific context depends on three antecedents: attitude toward the behavior,
subjective norm, and perceived behavioral control. Attitude toward the behavior (in our case: attitude
toward Internet adoption) reflects the individual’s own (positive or negative) evaluation of the behav-
ior in question. Subjective norms reflect an individual’s perception that “most people who are impor-
tant to him or her think s/he should or should not perform the behavior in question” (Fishbein and
Ajzen 1975: 302). Perceived behavioral control reflects an individual’s perception of his or her ability
to perform a given behavior.
All three antecedents of actual adoption behavior are context-dependent, suggesting that inhabi-
tants of neighboring countries tend to be, ceteris paribus, more similar with respect to attitudes, sub-
jective norms, and perceived behavioral control than people from more distant countries. That people
from neighboring countries often have very similar attitudes toward new technologies is well docu-
mented by cross-country surveys of attitudes and opinions toward science and technology in society
(e.g., the Eurobarometer surveys on EU citizens attitudes toward science and technology in general
(EU Commission 2014a,b) or the Special Eurobarometers on public attitudes toward Internet use and
data protection (EU Commission 2015), on biotechnology (EU Commission 2010) or on the use of
robots (EU Commission 2012). A similar argument holds for norms and values.9 As evidenced by
the World Values Survey and related research (see, e.g., Berggren and Nilsson 2015; Parts 2013;
Wach 2015), social norms and values tend to be clustered in groups of neighboring countries. There
is also some evidence of clustering across more distant countries that share a common history and
language (e.g., the group of Commonwealth countries), but spatial proximity doubtlessly facilitates
the spread of norms and values.10
The observed similarity of social norms and attitudes toward new technologies in neighboring
countries is, however, by no means coincidental. As is well-known, peer groups and reference per-
sons play an important role in the establishment of norms and in the forming of attitudes and inten-
tions (Falck, Heblich, and Luedemann 2012; Merton 1968; Venkatesh and Morris 2000).
Internet adoption might, however, not only depend on influential reference persons within a coun-
try. People might view the people of neighboring countries as a relevant peer or reference group. The
reference group theory states that a reference group may mean a group with which one compares
oneself in making a self judgement (Merton and Kitt 1950). Peer or reference groups might also
serve as role models for individual behavior, such as Internet adoption. Although reference groups
need not be restricted to neighboring countries, spatial proximity is clearly helpful as common cul-
tural and historical roots and common religious beliefs facilitate identification, and people have typi-
cally better knowledge of neighboring countries than of more distant countries. This is partly due to
the fact that international migration is—to a substantial part—migration between neighboring coun-
tries. Moreover, identity economics captures the idea that people make economic choices not only
based on monetary incentives, but also on their self-conception (Akerlof and Kranton 2000). Akerlof
“BAD NEIGHBORHOOD” AND INTERNET ADOPTION 245
and Snower (2016) hint at the important role of narratives in motivating human action. As the spread
of narratives is facilitated by geographic, cultural, and language proximity, it is plausible that neigh-
boring countries have the same (or rather similar) narratives that impact on norms, attitudes and
beliefs and—finally—on behavioral intentions and actual behavior.11
The upshot of our argument is that there are manifold interactions between neighboring countries
in the Internet diffusion process. The impact of neighboring countries might well go beyond mere
knowledge spillovers and take the form of cross-country spillovers of attitudes, norms, and beliefs
that impact on individual adoption behavior. This leads us to hypothesis 1:
H1 Internet adoption in a country is positively affected by Internet adoption in neighboring countries.
Urbanization. There are numerous studies suggesting that cities are the sites where economic,
cultural, and technical progress takes place (Duranton and Puga 2001; Feldman and Audretsch 1999;
Hall 1998). One might thus expect a positive relationship between urbanization and technology adop-
tion. This view reflects the so-called urban density theory, asserting that costs of Internet adoption
are decreasing in population density and size due to knowledge spillovers and the availability of
complementary infrastructure and inputs (Forman, Goldfarb, and Greenstein 2005). The urban den-
sity theory is, however, not undisputed. A competing theory, known as global village theory, empha-
sizes the unique function of the Internet breaking down communication barriers between
organizations, which implies that establishments in rural or small urban areas benefit most.12
As discussed in the literature review, empirical evidence on the role of urbanization is ambiguous, in
particular with respect to poor countries: There exist prominent studies (e.g., Andonova 2006; Crenshaw
and Robison 2006) hinting at a positive impact of urbanization on Internet adoption. There are, however,
other (and equally prominent) studies finding a negative impact (e.g., Chinn and Fairlie 2007, 2010;
Howard and Mazaheri 2009) or no significant impact at all (e.g., Wunnava and Leiter 2009).
In our view, this might be due to the fact that rapid urbanization in poor countries is a mixed
blessing. On the one hand, urbanization gives rise to agglomeration economies and can make Internet
adoption more attractive due to lower (per capita) provision costs and the availability of complemen-
tary infrastructure and inputs. On the other hand, many cities in poor countries have seen a rapid,
uncontrolled population growth in recent years, Today, among the approximately 500 cities in the
world with more than 1 million inhabitants, about 70 percent are cities in the developing world that
face very special problems, including massive congestion of infrastructure, shortage of housing, and
rapid expansion of “urban poor” settlements (UN, Department of Economic and Social Affairs, Pop-
ulation Division 2015). Fifty-five percent of the urban population in sub-Saharan Africa lives in
slums. The proportion is 31 percent in South Asia, 26 percent in East Asia and the Pacific, and 20
percent in Latin America (World Bank 2015). Such settlements often fall outside of the normal provi-
sion of public amenities including supplies of water, sanitation, sewage, and power (UNICEF 2014).
They face severe problems of overcrowding, pollution, crime, and extreme poverty (UN Department
of Economic and Social Affairs, Population Division 2015), which are definitely not conducive to
social and technical progress and likely to more than outweigh the advantages of urbanization.13
The variable most frequently used in the literature (the share of the population living in cities)
captures both, the upsides and downsides of urbanization in poor countries, and it is thus not surpris-
ing that results are ambiguous. We retain this variable (in order to ensure comparability with previous
work), but also run alternative models making use of a variable (the share of population living in
large cities with more than 1 million inhabitants) that better captures the peculiar problems of large,
rapidly growing urban agglomerations in poor countries. We expect the first measure (share of popu-
lation living in cities) to have an ambiguous impact and the second measure (share of population
246 GROWTH AND CHANGE, MARCH 2018
living in large agglomerations with more than one million inhabitants) to have a negative impact on
Internet adoption.
H2a The share of population living in cities (urbanization per se) has an ambiguous impact on Internet adoption in poor
countries.
H2b A high concentration of population in large cities (more than one million inhabitants) has a negative impact on Inter-
net adoption in poor countries.
Empirical Strategy and Data Methodology. We build on existing models of Internet adoption and incorporate a location vari-
able, neighbori,t-1, and an urbanization variable urbani,t-1 in the cross-country analysis of Internet
adoption. Our dependent variable, useri,t, is the percentage of Internet users in a country’s population.
Since technology diffusion is an accumulation of individual adoption decisions, the most relevant dif-
fusion measure for most technologies is the ratio of actual to potential users (Andres et al. 2010).14
The percentage of Internet users in a population is also one of the most widely used measures in the
literature (Andonova 2006; Beilock and Dimitrova 2003; Chinn and Fairlie 2007; Guill�en and Su�arez
2005). The variable Internet users is preferred over Internet subscribers and computer penetration
rates as it includes household access to the Internet, as well as Internet access from public places
such as universities, workplaces, and Internet cafes (Andre�s et al. 2010). As Internet adoption in
developing countries is only at a preliminary stage it typically follows a linear graph as can be seen
from Figure A1 in the Appendix. The graphs corroborate with the predictions of the S-shaped diffu-
sion curve, with the state of Internet adoption in developing countries concentrated at the initial linear
segment.
Our estimated model is as follows:
useri;t5 const 1 b1neighbori;t211 b2urbani;t211 c’Z 1 ai1 ei
where const is a constant applied to all observations, neighbori,t-1 is the macro-geographic location
(neighborhood) variable, urbani,t-1 is the urbanization variable, Z is the set of control variables and ai represents the unobserved heterogeneity (fixed effect) for each country. All independent variables
and control variables are lagged by 1 year.
The location variable neighbori,t-1 is the main variable of interest and is constructed as the lagged
average percentage of Internet users in neighboring countries. As discussed in the third section, we
consider two different measures of urbani,t-1. The first measure of urbanization is defined as the pro-
portion of the population living in cities (denoted as variable “urban”) and the second measure indi-
cates the proportion of population which lives in large agglomerations with more than 1 million
inhabitants (denoted as variable “urban million”).
The control variables included in our empirical model, represented by Z, follow the previous liter-
ature on Internet diffusion (Section 2). Standard measures used in prior work and considered as con-
trol in our estimations include measures of income (GDP per capita), telecommunications
infrastructure (telephone fixed lines), freedom of the press, years of schooling (as proxy for human
capital quality), and age structure of the population (considered by the population share of the
elderly). We consider three additional control variables: Exports of goods and services (as a percent-
age of GDP) is a measure of trade openness. The underlying idea is that countries which are more
engaged in international trade and exchange are more likely to adopt new technologies (in particular
such technologies that facilitate doing international business). This measure (or closely related
“BAD NEIGHBORHOOD” AND INTERNET ADOPTION 247
measures) has been used in several previous studies of Internet adoption across countries (e.g., Chinn
and Fairlie 2010; Crenshaw and Robison 2006). The share of the female labor force is a measure of
female labor market participation. Female labor market participation tends to be high in modern,
highly developed societies, and low in less-developed countries that discriminate women for religious
or other reasons. The expectation is that high female labor market participation rates tend to push
Internet adoption, whereas the exclusion of women from the formal labor market works in the oppo-
site direction. Time to business is a relatively new measure provided by the World Bank that mea-
sures the ease of doing business and (more generally) the business-friendliness of the institutional
environment in a country. As our dependent variable is total Internet adoption (which includes pri-
vate as well as business users) we expect a business-friendly institutional environment to have a posi-
tive influence on total Internet usage.15 We further control for unobserved, not time-varying
heterogeneity across countries by estimating country fixed effects. A detailed description of all varia-
bles used in the empirical analysis is provided in Table A1 in the Appendix. Table A2 displays the
descriptive summary statistics.
An important assumption implicit in our estimation is the exogeneity of the regressors. While we
have lagged the neighborhood variable by one time period, we cannot rule out the possibility of
endogeneity and reverse causation. To take care of potential endogeneity of the neighborhood vari-
able and allow for consistent estimation we perform instrumental variable (IV) regressions. We use
past adoption of telephones—as measured by the average share of telephone fixed line subscribers
from 1960 to 1995 in neighboring countries—as an instrument for Internet adoption in neighboring
countries today, and denote this variable as “neighbor (IV).” The rationale for this choice of instru-
ment is that countries that were quick to adopt major technological innovations in the past are likely
to be quick in adopting current innovations as well. Past telephone adoption is well-suited for this
purpose as it is well documented in the literature (e.g., Brooks 1976) that developing countries were
at the initial stages of telephone adoption in the 1960s, similar to Internet adoption levels around
2005. While we can only instrument telephone subscribers in five year intervals due to data con-
straints, past variations in telephone subscribers is a highly relevant instrument for variations in Inter-
net users today as evidenced by the results of the instrument relevance test displayed in Table A4 in
the appendix.16 The instrument (telephone subscribers in the past) is obviously not influenced by
Internet adoption today, and the correlation between our main explanatory variable (the average share
of Internet users in neighboring countries) and the instrumental variable (the average share of fixed
line telephone subscribers in neighboring countries in previous periods) is high (above 73 percent).
Hence, the chosen instrument satisfies the two key requirements (exogeneity and high correlation
with the endogenous regressor) that constitute a valid instrument (e.g., Cameron and Trivedi 2005;
Davidson and MacKinnon 2004; Wooldridge 2010).
Data. Panel data on 63 developing countries on telecommunications and technology usage and
country demographic and institutional characteristics were compiled for the years 2005–2013. We
follow the World Bank Country Classifications17 in defining developing countries as low income
and low-middle income countries. The full list of countries considered in our sample is displayed in
Table A3 in the Appendix. Data on telecommunications and technology use are obtained from the
International Telecommunication Union. The main source of socioeconomic data, such as income,
country demographics, and institutions, is the World Development Indicators database by the World
Bank. We also use indices of freedom of the press from Freedom House as a proxy for political free-
dom. Our primary measure of human capital, average years of schooling, is derived from the
UNESCO Institute for Statistics. More details of the variables can be found in Tables A1 and A2 in
the Appendix.
248 GROWTH AND CHANGE, MARCH 2018
As can be seen from Table 2, correlation between most explanatory variables is low, although a
few control variables display levels of correlation larger than 0.5 with some of the other variables.
We thus performed variance inflation tests and ran additional regressions excluding variables that
drive up variance inflation, finding that this had no impact on the main results (reported in the fol-
lowing section).
Findings Table 3 displays the results of our baseline regression (with and without controls) and with two
different urbanization measures. Models (2)–(4) use the traditional urbanization variable (share of
population living in cities), whereas models (5)–(7) use “urban million” (share of population in large
cities with more than 1 million inhabitants). The differences between the models in Table 3 are as
follows: Model (1) is the model without controls. In models (2) and (5), the urbanization variable
(“urban” in model (2) and “urban million” in model (5)) is added. Models (3) and (6) use a limited
set of control variables,18 whereas models (4) and (7) use the full set of control variables.
The key finding is that Internet adoption in low and lower middle-income countries is positively
affected by adoption rates in neighboring countries. This result is robust for different model specifica-
tions and also holds when we control for a rich set of other potential determinants of Internet adop-
tion (Table 3).19
Urbanization (as measured by the share of population living in cities) has an ambiguous impact
on Internet adoption: The coefficient is positive (and weakly significant) in the model without con-
trols, and insignificant in the models that consider controls. Using our preferred measure of urbaniza-
tion in poor countries (“urban million”) clearly increases the fit of the model,20 but the urbanization
variable remains insignificant (models 5–7 in Table 3).21
While our results hint at a significant positive correlation of Internet adoption rates in neighboring
countries, a causal interpretation is not possible as causality might run in both directions. In order to
tackle the endogeneity problem, we make use of an instrumental variable estimation. We analyzed
Internet adoption in neighboring countries with past telephone adoption in neighboring countries. As
discussed in Methodology section, the average share of telephone fixed line subscribers in neighbor-
ing countries (in 5-year intervals from 1960 to 1995) is a valid variable, as it is clearly exogenous
and highly correlated with our main variable of interest.22
As can be seen from Table 4, the results of the instrumental variable estimation confirm the results
of the baseline model. We consider this strong evidence for the role of macro-geographic location
(neighborhood) in determining Internet adoption rates in less developed countries. As discussed in
the third section, there are good theoretical arguments for an important role of neighboring countries.
‘Good neighbors’ create positive spillovers, and it is important to note that such positive spillovers
can go beyond knowledge spillovers in a narrow sense and also take the form of spillovers of atti-
tudes and beliefs, brought about by role models from neighboring countries or the spread of social
norms and narratives (characterizing risk-taking and technical progress as positive values) from
neighboring countries.
Our results, displayed in Table 4, are in line with recent anecdotal evidence provided by the
World Bank: An example of a “good neighborhood” noted by the World Bank is the group of neigh-
boring post-Soviet states, Kazakhstan, Kyrgyz Republic, Tajikistan, and Uzbekistan. In recent years,
there seems to be a contagion effect in promoting ICT infrastructure in this region as neighboring
countries began rolling out national infrastructure plans23 successively after each other (World Bank
2016b). A “bad neighborhood,” by contrast, is characterized by a lack of positive spillovers and a
“BAD NEIGHBORHOOD” AND INTERNET ADOPTION 249
T A
B L
E 2
. C
O R
R E
L A
T IO
N M
A T
R IX
.
1 2
3 4
5 6
7 8
9 1
0 1
1 1
2 1
3
1 N
ei g
h b
o r
1
2 N
ei g
h b
o r
(I V
) 0
.7 3
1 1
3 U
rb an
0 .4
6 5
0 .3
7 6
1
4 U
rb an
m il
li o
n 0
.3 2
6 0
.3 2
0 0
.7 3
5 1
5 T
el fi
x ed
li n
es 0
.5 9
8 0
.5 4
7 0
.6 3
4 0
.3 7
2 1
6 G
D P
_ ca
p 0
.1 8
4 0
.1 9
8 0
.0 3
9 0
.0 0
9 0
.1 1
8 1
7 S
ch o
o li
n g
0 .5
1 4
0 .4
8 9
0 .6
2 1
0 .4
6 2
0 .7
1 8
0 .0
6 7
1
8 F
re ed
o m
o f
p re
ss 2
0 .0
6 7
2 0
.0 1
0 2
0 .2
6 1
2 0
.1 9
0 2
0 .0
1 1
2 0
.0 4
4 0
.0 6
4 1
9 T
im e
to b
u si
n es
s 0
.2 6
0 2
0 .0
9 4
2 0
.1 2
0 2
0 .0
6 3
2 0
.2 1
6 2
0 .0
2 6
2 0
.0 6
1 0
.0 8
9 1
1 0
E x
p o
rt 0
.1 1
7 0
.0 0
1 0
.2 0
5 0
.1 2
3 0
.2 3
5 0
.0 0
5 0
.2 4
5 0
.1 5
0 0
.2 2
9 1
1 1
L ab
o r
fe m
al e
2 0
.2 1
2 2
0 .2
8 9
2 0
.3 7
2 2
0 .3
0 1
2 0
.3 1
0 2
0 .2
4 6
2 0
.0 8
3 2
0 .0
1 2
0 .3
0 1
0 .0
9 0
1
1 2
U n
em p
lo y
m en
t 0
.0 8
9 0
.2 3
1 0
.2 2
3 0
.1 6
1 0
.2 5
1 2
0 .0
2 4
0 .3
1 5
2 0
.0 3
8 2
0 .1
9 4
2 0
.1 7
3 2
0 .1
9 0
1
1 3
O ld
er th
an 6
4 0
.6 1
0 0
.5 1
2 0
.5 9
2 0
.3 5
1 0
.8 9
7 0
.0 9
8 0
.7 6
9 2
0 .0
4 4
2 0
.1 5
6 0
.2 0
7 2
0 .1
7 8
0 .3
1 0
1
250 GROWTH AND CHANGE, MARCH 2018
dominance of negative spillovers. Sub-Saharan Africa is a typical example of a “bad neighborhood,”
with low income per capita, underdeveloped production structures, and poor human capital and phys-
ical infrastructure investments (World Bank 2016b), and it is notable that African countries such as
Egypt and Swaziland have experienced slower growth in Internet adoption from 2005 to 2013
despite having higher GDP per capita than Bolivia and Guyana in South America. Although the
econometric evidence indicating a strong neighborhood effect is very robust, a potential limitation of
our country-level analysis needs to be mentioned as well. Country-level analyses implicitly assume
that nations are homogenous entities. In fact, however, there is often substantial geographic variation
of Internet use within nations. This has been shown for highly developed countries like the U.S.
(Warf 2013) as well as for developing economies like China, Mexico, or Indonesia (Li and Shiu
TABLE 3. DETERMINANTS OF INTERNET ADOPTION RATES IN LOW AND LOWER MIDDLE-INCOME COUN- TRIES 2005–2013 (BASELINE ESTIMATIONS).
(1) (2) (3) (4) (5) (6) (7)
User User User User User User User
Neighbor 0.830*** 0.753*** 0.436*** 0.401* 0.939*** 0.598*** 0.871***
(0.142) (0.160) (0.129) (0.222) (0.205) (0.123) (0.207)
Urban 0.692* 0.190 20.0942
(0.388) (0.310) (0.371)
Urban million 0.504 20.0795 20.614
(0.807) (0.575) (0.539)
Tel fixed lines 0.0483 0.0951
(0.170) (0.215)
GDP_ cap 0.0062*** 0.0076*** 0.0064*** 0.0050**
(0,00116) (0.00218) (0.00169) (0.00245)
Schooling 0.930 0.737 1.021 0.722
(1.328) (1.482) (1.350) (1.920)
Freedom of press 0,0556 0.0694 0.105 0.177*
(0,0651) (0.0722) (0.0804) (0.0984)
Time to business 0.0108 0.00657
(0.00862) (0.0180)
Export 0.0623 0.103
(0.0574) (0.0714)
Labor_female 0.576 0.197
(0.605) (0.537)
Unemployment 20.0717 20.139
(0.274) (0.283)
Older than 64 24,740 22.175 26.885 23.569
(2,715) (2.844) (3.584) (3.206)
_cons 20.903 225.48* 211,29 250.56 29.542 1.920 219.20
(1.749) (13.36) (15,27) (30.52) (11.12) (19.65) (31.51)
N 464 464 452 382 305 297 247
adj. R2 0.563 0.579 0,712 0.744 0.598 0.752 0.814
Standard errors in parentheses *p < 0.10; **p < 0.05; ***p < 0.01.
“BAD NEIGHBORHOOD” AND INTERNET ADOPTION 251
2012; Freedom House 2015; Sujarwoto and Tampubolon 2016). Hence, it cannot be excluded that a
country may be bordering on a neighboring nation’s sub-region that is much higher in Internet use or
much lower in Internet use than that neighbor’s average Internet use.24
Table 4 also shows a slight change in the results for the urbanization variables. The standard
urbanization variable (“urban”) is no longer significant, whereas our preferred urbanization vari-
able (“urban million”) has a negative sign in all model specifications, and is (weakly) significant
in the model considering the full set of control variables (column 7). It should be noted that a
negative sign of the “urban million” variable does not necessarily imply that Internet adoption in
large agglomerations is lower than in rural areas. If differences in the total share of urban popu-
lation across countries are not too large, a negative sign of the “urban million” variable might
TABLE 4. DETERMINANTS OF INTERNET ADOPTION RATES IN LOW AND LOWER MIDDLE-INCOME COUNTRIES 2005–2013 (IV ESTIMATIONS).
(1) (2) (3) (4) (5) (6) (7)
User User User User User User User
Neighbor (IV) 1.084*** 1.081*** 0.763*** 0.928*** 1.25*** 0.817*** 1.225***
(0.0505) (0.0607) (0.141) (0.204) (0.075) (0.147) (0.158)
Urban 0.0367 0.0079 20.204
(0.211) (0.215) (0.239)
Urban million 20.411 20.170 20.583*
(0.468) (0.335) (0.305)
Tel fixed lines 20.0649 20.0273
(0.116) (0.130)
GDP_cap 0.00456*** 0.00360** 0.00500*** 0.00216
(0.00119) (0.00183) (0.00122) (0.00156)
Schooling 21.543 21.743 20.482 20.862
(1.292) (1.368) (1.244) (1.224)
Freedom of press 0.0453 0.0791 0.108** 0.188***
(0.0419) (0.0500) (0.0504) (0.0681)
Time to business 0.0172** 0.0127
(0.00694) (0.0152)
Export 0.0772* 0.106**
(0.0446) (0.0467)
Labor_female 0.478 0.156
(0.339) (0.300)
Unemployment 0.0654 20.103
(0.357) (0.245)
Older than 64 23.228 21.114 26.474*** 23.124*
(1.763) (1.751) (1.800) (1.798)
N 437 425 353 300 292 237
adj. R2 0.521 0.615 0.629 0.579 0.706 0.773
Standard errors in parentheses *p < 0.10; **p < 0.05; ***p < 0.01.
252 GROWTH AND CHANGE, MARCH 2018
just result from the fact that countries with a more balanced urban structure (including smaller
cities below 1 million inhabitants as well as larger cities) tend to have higher Internet adoption
rates than countries in which all urban population is concentrated in a small number of very
large cities. Given the well-known problems of excessively growing megacities in the poorest
countries (including extreme poverty, crime, social exclusion, breakdown of critical infrastruc-
ture, etc.) the latter interpretation appears quite plausible. Moreover, as rapid urban growth in
the developing countries has outstripped the capacity of many of the largest cities to supply
basic services for their inhabitants, the pressures and problems of urban congestion are often
more severe in more populated cities, i.e., with a population of more than one million. Although
this is one possible (and likely) explanation for the negative correlation of “urban million” with
Internet adoption, urbanization results should not be over-interpreted, as “urban million” is only
significant at the 90 percent confidence level and in one model specification. Further research
taking into account urban structure (and not only urbanization rates per se) is needed to better
understand the complex role of urban structure for Internet adoption in poor countries.
Conclusion and Outlook Within the rich literature dealing with the determinants of Internet adoption and the digital divide,
a pertinent factor, macro-geographic location, has received relatively little attention so far. A key
finding of the current paper is that Internet adoption in low and lower middle income countries is
positively affected by adoption rates in neighboring countries, even when controlling for a wide
range of covariates (such as per capita income, education or institutions). While “good neighbors”
create positive spillovers, and push Internet adoption, many of the poorest countries appear to suffer
from “bad neighborhood,” characterized by a lack of positive and a dominance of negative spillovers.
Own efforts of the poorest countries to improve their institutional settings, their education system,
and their analog and digital infrastructure are necessary, but they are unlikely to be sufficient. As the
World Bank concluded in its current World Development Report, “for digital technologies to benefit
everyone everywhere requires closing the remaining digital divide, especially in Internet access,”
such that international policy action is indispensable (World Bank 2016a). Our results imply that
international policies to support Internet adoption in poor countries might be more effective if they
target groups of neighboring countries rather than single countries, as they can better exploit spill-
overs between neighboring countries. The same argument holds for development aid that might be
more effective if it takes a broader view, and carefully takes the strong interrelation between Internet
adoption in neighboring countries evidenced by this paper into account. Moreover, appropriate poli-
cies to address the “digital divide” should not only focus on availability and endowment (i.e.,
“hardware” aspects), but also provide incentives for an interactive and creative use of the Internet
throughout society (Camagni and Capello 2005).
Our findings have important implications for research as well. The different forms of spillovers
between neighboring countries are—due to their intangible nature—not well explored as yet. The analy-
sis in the third section suggests that they might go well beyond mere knowledge spillovers and take the
form of cross-country spillovers of attitudes and beliefs brought about by role models from neighboring
countries or the spread of social norms and narratives, which is clearly facilitated by spatial proximity.
These forms of cross-country spillovers in technology adoption are not well captured by mainstream eco-
nomics as yet. New approaches—such as identity economics—might be helpful in establishing a broader
theoretical basis, allowing scholars to analyze spillovers between neighboring countries more
thoroughly.
“BAD NEIGHBORHOOD” AND INTERNET ADOPTION 253
NOTES 1. The developed/developing country classifications are based on the UN M.49 standard.
2. The digital gap is measured as difference in the shares of Internet adopters in developed and developing countries.
3. There is no exact information about the observation period in the paper. It is only mentioned that the data used were pub-
lished in 2000.
4. Chinn and Fairlie (2010) use the same 161 country sample as Chinn and Fairlie (2007), but update it to the 2002–2004
period. They use the Blinder-Oaxaca technique to explain the gap in computer and Internet penetration between developed
and developing countries.
5. The studies by Oyelaran-Oyeyinka and Lal (2005) and Chinn and Fairlie (2010) are notable exceptions. They are, however,
a bit outdated (Oyelaran-Oyeyinka and Lal 2005, relates to the period 1995–2000 and Chinn and Fairlie 2010, relates to
2002–2004), and they do not consider neighborhood effects.
6. Unlike other studies on cross-country Internet adoption that take location into account (e.g., Forman, Goldfarb, and Green-
stein 2005), we are not focusing on commercial Internet diffusion, but aggregate Internet usage across countries.
7. Keller finds that the distance at which the amount of spillovers is halved is about 1,200 km.
8. Liaw (2004), Brown and Venkatesh (2005), Pedersen (2005).
9. Neighboring countries often share a common culture and language and a common (colonial) history. They tend to have
similar religious and ethical beliefs and values and a similar view of the individual in society (individual versus collectivist
societies).
10. We are not aware of cross-country data for perceived behavioral control (reflecting such factors as self-efficacy and
risk preference, but also resource and skill constraints that reduce a person’s ability to perform a given behavior). It
appears likely, however, that average perceived behavioral control tends to be similar in neighboring countries.
Clearly, personal traits like self-efficacy and risk preference are not exogenous, but depend on the social environment
in which a person grows up. As argued before, social norms and social environments tend to be more similar in neigh-
boring than in distant countries. It is also plausible that the prevalence of restricting factors (resource and skill contra-
ints) is similar in neighboring countries, as they are often at a similar development level and face similar social
problems.
11. There is also a literature on “social movement spillover,” focused on how new social movements are influenced by exist-
ing movements and borrow from their strengths and strategies (Meyer and Whittier 1994). The events of the “Arab
Spring” provide evidence that social movements do not only spill over to other movements, but also to neighboring
countries.
12. The global village theory is thus better suited for commercial Internet adoption than for total Internet adoption that includes
the adoption behavior of private households.
13. Those who might benefit most from the Internet often have the least chance to use it. “Indeed, the ‘information-poor’ are
typically unaware of the massive economic, technological and political changes that exclude them further from the ‘infor-
mation society’ . . .” (Warf 2001: 12).
14. Other Internet adoption measures in the literature include host counts, number of computers, broadband subscribers.
15. We also control for unemployment, expecting that high unemployment rates tend to decrease Internet usage.
16. Average telephone fixed lines subscribers in neighboring countries from 1960 to 1995 has an instrumental relevance test of
F 5 69.70 (model with full set of controls) and F 5 38.48 (model with reduced set of controls). Both are highly significant.
See Table A4 in the Appendix for details.
17. The World Bank classifies countries into low-income, lower-middle income, upper-middle, and high-income groups according
to their GNI per capita. As of June 2016, low-income economies are defined as those with a GNI per capita of $1,045 or less,
whereas lower middle-income economies are those with a GNI per capita of more than $1,045 but less than $4,125.
18. The “limited set of controls”-models (3) and (6) exclude variables that drive up variance inflation or were less often used in
previous work.
19. The model with the best fit (adj R2 5 0.814) is the one using the full set of control variables and using “urban million” as
urbanization variable.
20. The adjusted R2 increases from 0.744 to 0.814 in the models that consider the full set of control variables.
21. The only significant controls are GDP per capita and freedom of the press. Both have the expected signs.
22. First stage results of the IV estimation and results of the instrumental variable relevance test are shown in Table A4 in the
Appendix.
254 GROWTH AND CHANGE, MARCH 2018
23. Some examples of national infrastructure plans focusing on promoting ICT infrastructure are “Digital Kazakhstan 2020” in
Kazakhstan, “Digital Kyrgyzstan 2020–2025” in Kyrgyz Republic, “National Development Strategy 2030” in Tajikistan, and
“ICT Infrastructure Development Program 2015–2019” and “E-Government Development Program 2013–2020” in Uzbekistan.
24. We are grateful to an anonymous referee for hinting at this point.
REFERENCES Ajzen, I. 1991. The theory of planned behavior. Organizational Behavior and Human Decision Processes 50(2): 179–211.
Akerlof, G., and R. Kranton. 2000. Economics and identity. Quarterly Journal of Economics 105(3): 715–753.
Akerlof, G., and D. Snower. 2016. Bread and bullets. Journal of Economic Behavior & Organization 126: 58–71.
Andonova, V. 2006. Mobile phones, the Internet and the institutional environment. Telecommunications Policy 30(1): 29–45.
Andre�s, L., D. Cuberes, M. Diouf, and T. Serebrisky. 2010. Diffusion of the Internet: A cross-country analysis.
Telecommunications Policy 34(5–6): 323–340.
Beilock, R., and D. Dimitrova. 2003. An exploratory model of inter-country Internet diffusion. Telecommunications Policy
27(3–4): 237–252.
Berggren, N., and T. Nilsson. 2015. Globalization and the transmission of social values: The case of tolerance. Journal of
Comparative Economics 43(2): 371–389.
Brooks, J. 1976. Telephone: The first hundred years. Scranton, PA: HarperCollins.
Brown, S., and V. Venkatesh. 2005. Model of adoption of technology in households: A baseline model test and extension
incorporating household life cycle. MIS Quarterly 29(3): 399–426.
Camagni, R., and R. Capello. 2005. ICTs and territorial competitiveness in the era of Internet. Annals of Regional Science
39(3): 421–438.
Cameron, A., and P. Trivedi. 2005. Microeconometrics: Methods and applications. Cambridge: Cambridge University Press.
Capello, R., and P. Nijkamp. 1996a. Information and communications networks in space: Introduction to the special issue.
Annals of Regional Science 30(1): 1–5.
———. 1996b. Telecommunications technologies and regional development: Theoretical considerations and empirical evi-
dence. Annals of Regional Science 30(1): 7–30.
Caselli, F., and W. Coleman. 2001. Cross-country technology diffusion: The case of computers. American Economic Review
91(2): 328–335.
Chinn, M., and R. Fairlie. 2007. The determinants of the global digital divide: Across-country analysis of computer and Inter-
net penetration. Oxford Economic Papers 59(1): 16–44.
———. 2010. ICT use in the developing world: An analysis of differences in computer and Internet penetration. Review of
International Economics 18(1): 153–167.
Choucri, N., S. Madnick, and J. Ferwerda. 2014. Institutions for cyber security: International responses and global imperatives.
Information Technology for Development 20(2): 96–121.
Collier, P. 2007. The bottom billion. Why the poorest countries are failing and what can be done about it. Oxford: Oxford
University Press.
Comin, D., and M. Mestieri. 2013. Technology diffusion: Measurement, causes and consequences. NBER Working Paper
19052. Cambridge, MA: National Bureau of Economic Research.
Crenshaw, E., and K. Robison. 2006. Globalization and the digital divide: The roles of structural conduciveness and global
connection in Internet diffusion. Social Science Quarterly 87(1): 190–207.
Cruz-Jesus, F., T. Oliveira, and F. Bacao. 2012. Digital divide across the European Union. Information & Management 49(6):
278–291.
Cruz-Jesus, F., M. Vicente, F. Bacao, and T. Oliveira. 2016. The education-related digital divide: An analysis for the EU-28.
Computers in Human Behavior 56: 72–82.
Davidson, R., and J. MacKinnon. 2004. Econometric theory and methods, Vol.5. New York: Oxford University Press.
Davison, R., D. Vogel, R. Harris, and N. Jones. 2000. Technology leapfrogging in developing countries-an inevitable luxury?
Electronic Journal of Information Systems in Developing Countries. 1(2000): 1–10. www.ejisdc.org
DiMaggio, P., E. Hargittai, W. Neuman, and J. Robinson. 2001. Social implications of the Internet. Annual Review of Sociol-
ogy 27(1): 307–336.
Dohse, D., and C.Y. Lim. 2016. Macro-geographic location and Internet adoption in poor countries: What is behind the per-
sistent digital gap? Kiel Working Paper 2067, Kiel Institute for the World Economy, Kiel, Germany.
“BAD NEIGHBORHOOD” AND INTERNET ADOPTION 255
Duranton, G., and D. Puga. 2001. Nursery cities: Urban diversity, process innovation, and the life cycle of products. American
Economic Review 91(5): 1454–1477.
EU Commission. 2010. Europeans and biotechnology in 2010: Winds of change? Brussels: European Commission.
———. 2012. Public Attitudes towards Robots. Special Eurobarometer 382. Brussels: European Commission.
———. 2014a. Social climate and innovation in science and technology. Eurobarometer 81.5 (ICPSR 36241). Brussels: Euro-
pean Commission.
———. 2014b. Public perceptions of science, research and innovation. Special Eurobarometer 419. Brussels: European
Commission.
———. 2015. Europeans in 2015, Data protection and internet use. Eurobarometer 83.1. Brussels: European Commission.
Falck, O., S. Heblich, and E. Luedemann. 2012. Identity and entrepreneurship: do school peers shape entrepreneurial inten-
tions? Small Business Economics 39(1): 39–59.
Feldman, M., and D. Audretsch. 1999. Innovation in cities: Science-based diversity, specialization and localized competition.
European Economic Review 43(2): 409–429.
Fishbein, M., and I. Ajzen. 1975. Belief, attitude, intention and behavior: An introduction to theory and research. Reading,
MA: Addison-Wesley.
Forman, C., A. Goldfarb, and S. Greenstein. 2005. Geographic location and the diffusion of Internet technology. Electronic
Commerce Research and Applications 4: 1–13.
Freedom House. 2015. Freedom on the Net: Country report Mexico. https://freedomhouse.org/report/freedom-net/2015/mexico
(accessed July 15, 2016).
Goldfarb, A., and J. Prince. 2008. Internet adoption and usage patterns are different: Implications for the digital divide. Infor-
mation Economics and Policy 20(1): 2–15.
Guill�en, M., and S. Su�arez. 2005. Explaining the global digital divide: Economic, political and sociological drivers of cross-
national Internet use. Social Forces 84(2): 681–708.
Hall, P. 1998. Cities in civilization: Culture, technology, and urban order. New York: Pantheon.
Hargittai, E. 1999. Weaving the Western web: Explaining differences in Internet connectivity among OECD countries. Tele-
communications Policy 23(10–11): 701–718.
Hirt, M., and P. Willmot. 2014. Strategic principles for competing in the digital age. McKinsey Quartely, http://www.mckin-
sey.com/business-functions/strategy-and-corporate-finance/our-insights/strategic-principles-for-competing-in-the-digital-age
(accessed July 15, 2016).
Howard, P., and N. Mazaheri. 2009. Telecommunications reform, Internet use and mobile phone adoption in the developing
world. World Development 37(7): 1159–1169.
Keller, W. 2002. Geographic localization of international technology diffusion. American Economic Review 92(1): 120–142.
Kiiski, S., and M. Pohjola. 2002. Cross-country diffusion of the Internet. Information Economics and Policy 14(2): 297–310.
Li, R., and A. Shiu. 2012. Internet diffusion in China: A dynamic panel data analysis. Telecommunications Policy 36(10): 872–887.
Liaw, S. 2004. The theory of planned behaviour applied to search engines as a learning tool. Journal of Computer Assisted
Learning 20(4): 283–291.
Merton, R. 1968. Social theory and social structure. New York: Free Press.
Merton, R., and A. Kitt. 1950. Contributions to the theory of reference group behavior. Glencoe: Free Press.
Meyer, D., and N. Whittier. 1994. Social movement spillover. Social Problems 41(2): 277–298.
Niehaves, B., and R. Plattfaut. 2014. Internet adoption by the elderly: Employing IS technology acceptance theories for under-
standing the age-related digital divide. European Journal of Information Systems 23(6): 708–726.
Norris, P. 2001. Digital divide: Civic engagement, information poverty, and the Internet worldwide. Cambridge: Cambridge
University Press.
Oyelaran-Oyeyinka, B., and K. Lal. 2005. Internet diffusion in sub-Saharan Africa: A cross-country analysis. Telecommunica-
tions Policy 29(7): 507–527.
Oyeleye, O., M. Sanni, and T. Shittu. 2015. An investigation of the effects of customers’ educational attainment on their adop-
tion of e-banking in Nigeria. Journal of Internet Banking and Commerce 20(3): 1–16.
Parts, E. 2013. Social capital, national values and attitudes towards immigrants: Empirical evidence from the European Union
and neighbouring countries. SEARCH Working Paper 3/19. University of Tartu, Estonia.
Pedersen, P. 2005. Adoption of mobile Internet services: An exploratory study of mobile commerce early adopters. Journal of
Organizational Computing and Electronic Commerce 15(2): 203–221.
256 GROWTH AND CHANGE, MARCH 2018
P�enard, T., N. Poussing, B. Mukoko, and G. Piaptie. 2015. Internet adoption and usage patterns in Africa: Evidence from
Cameroon. Technology in Society 42: 71–80.
Priluck, R. 2016. Internet diffusion and adoption in Cuba. Atlantic Marketing Journal 5(2): 1–18.
Rodriguez, F., and E. Wilson. 2000. Are poor countries losing the information revolution? InfoDev Working Paper 26651.
Washington, DC: World Bank.
Sujarwoto, S., and G. Tampubolon. 2016. Spatial inequality and the Internet divide in Indonesia 2010–2012.
Telecommunications Policy 40(7): 602–616.
Tengtrakul, P., and J. Peha. 2013. Does ICT in schools affect residential adoption and adult utilization outside schools?
Telecommunications Policy 37(6): 540–562.
Touray, A., A. Salminen, and A. Mursu. 2015. Internet adoption at the user level: Empirical evidence from the Gambia.
Information Technology for Development 21(2): 281–296.
UNICEF. 2014. Phnom Penh: Multiple indicator assessment of the urban poor. http://www.unicef.org/cambodia/PIN_
URBAN_POOR_FA.PDF (accessed May 10, 2016).
UN, Department of Economic and Social Affairs, Population Division. 2015. World urbanization prospects: The 2014 revision
(ST/ESA/SER.A/336). New York: United Nations.
US Department of Commerce. 2002. A nation online: How Americans are expanding their use of the Internet. Washington,
DC: US Government Printing Office.
Venkatesh, V., and M. Morris. 2000. Why don’t men ever stop to ask for directions? Gender, social influence, and their role in
technology acceptance and usage behavior. MIS Quarterly 24(1): 115–139.
Wach, K. 2015. Impact of cultural and social norms on entrepreneurship in the EU: Cross-country evidence based on GEM
survey results. Zarządzanie w Kulturze 16(1): 15–29.
Wallsten, S. 2005. Regulation and Internet use in developing countries. Economic Development and Cultural Change 53(2):
501–523.
World Bank. 2015. World development indicators: Population living in slums (% of urban population). http://data.worldbank.
org/indicator/EN.POP.SLUM.UR.ZS (accessed November 15, 2016).
———. 2016a. Digital dividends. World Development Report 2016. Washington, DC: The World Bank Group.
———. 2016b. Reaping the benefits of digital technology in Central Asia. http://www.worldbank.org/en/news/feature/2016/
03/15/reaping-the-benefits-of-digital-technology-in-central-asia (accessed November 15, 2016).
Warf, B. 2001. Segueways into cyberspace: Multiple geographies of the digital divide. Environment and Planning B: Planning
and Design 28(1): 3–19.
———. 2009. The rapidly evolving geographies of the Eurasian Internet. Eurasian Geography and Economics 50(5): 564–580.
———. 2013. Contemporary digital divides in the United States. Tijdschrift Voor Economische En Sociale Geografie 104(1): 1–17.
West, D. 2015. Digital divide: Improving Internet access in the developing world through affordable services and diverse con-
tent. http://www.brookings.edu/~/media/research/files/papers/2015/02/13 (accessed November 20, 2016).
Wooldridge, J. 2010. Econometric analysis of cross section and panel data. Cambridge, MA: MIT Press.
Wunnava, P., and D. Leiter. 2009. Determinants of inter-country Internet diffusion rates. American Journal of Economics and
Sociology 68(2): 413–426.
“BAD NEIGHBORHOOD” AND INTERNET ADOPTION 257
Appendix
TABLE A1. VARIABLE DESCRIPTION AND DATA SOURCES.
Variable Description Source
User Percentage of Internet users in a country ITU
Neighbor Average percentage of Internet users in
neighboring countries
ITU
Neighbor (IV) Average percentage of telephone fixed line
subscribers from 1960 to 1995 (in 5-year-
intervals) in neighboring countries
ITU
Urban Urban population (% of total population) World Bank
Urban million Urban population in agglomerations of more
than 1 million (% of total population)
World Bank
Tel fixed lines Telephone fixed-lines per population ITU
GDP_cap GDP per capita (PPP) World Bank
Schooling Average years of schooling UNDP
Freedom of press Freedom of Press Ratings Freedom House
Time to business Number of calendar days needed to set up a business World Bank
Export Exports of goods and services (% of GDP) World Bank
Labor female Female Labor force (% of total Labor force) World Bank
Unemployment Total unemployment (% of total Labor force) World Bank
Older than 64 Population aged over 64 years old (% of total population) World Bank
TABLE A2. DESCRIPTIVE SUMMARY STATISTICS.
Variable Obs Mean Std. dev. Min Max
User 557 8.78348 9.966485 0.065239 56
Neighbor 472 12.17857 11.72404 0.255 57
Neighbor (IV) 445 2.191323 3.571777 0.0319996 26.80672
Urban 567 36.80471 14.61583 9.375 69.274
Urban million 369 15.25189 8.519043 3.108902 46.82812
Tel fixed lines 554 4.852888 6.240416 0 30.64515
GDP_cap 558 3261.009 2320.175 530.9611 10580.9
Schooling 562 5.055872 2.643623 1.3 12.1
Freedom of press 567 58.903 16.7821 24 97
Time to business 519 37.9264 37.3605 2 260
Export 526 31.27736 15.39206 5.51685 87.06688
Labor_female 567 57.5381 18.38306 13.9 87.4
Unemployment 567 7.791711 6.397262 0.1 36.4
Older than 64 567 4.262633 2.472458 2.176046 16.13981
258 GROWTH AND CHANGE, MARCH 2018
TABLE A3. LIST OF DEVELOPING COUNTRIES IN THE SAMPLE.
Country Country code World bank classification
Afghanistan AFG Low income
Armenia ARM Lower Middle Income
Bangladesh BGD Lower Middle Income
Benin BEN Low income
Bhutan BTN Lower Middle Income
Bolivia BOL Lower Middle Income
Burkina Faso BFA Low income
Burundi BDI Low income
Cambodia KHM Low income
Cameroon CMR Lower Middle Income
Cape Verde CPV Lower Middle Income
Chad TCD Low income
Central African Republic CAF Low income
Comoros COM Low income
Côte d’Ivoire CIV Lower Middle Income
Egypt EGY Lower Middle Income
El Salvador SLV Lower Middle Income
Ethiopia ETH Low income
Gambia GMB Low income
Georgia GEO Lower Middle Income
Ghana GHA Lower Middle Income
Guatemala GTM Lower Middle Income
Guinea GIN Low income
Guinea-Bissau GNB Low income
Guyana GUY Lower Middle Income
Haiti HTI Low income
Honduras HND Lower Middle Income
India IND Lower Middle Income
Indonesia IDN Lower Middle Income
Kyrgyzstan KGZ Lower Middle Income
Lao PDR LAO Lower Middle Income
Lesotho LSO Lower Middle Income
Liberia LBR Low income
Madagascar MDG Low income
Malawi MWI Low income
Mali MLI Low income
Mauritania MRT Lower Middle Income
Morocco MAR Lower Middle Income
Mozambique MOZ Low income
Myanmar MMR Lower Middle Income
Nepal NPL Low income
“BAD NEIGHBORHOOD” AND INTERNET ADOPTION 259
TABLE A3. CONTINUED
Country Country code World bank classification
Nicaragua NIC Lower Middle Income
Niger NER Low income
Nigeria NGA Lower Middle Income
Pakistan PAK Lower Middle Income
Papua New Guinea PNG Lower Middle Income
Philippines PHL Lower Middle Income
Rwanda RWA Low income
Senegal SEN Lower Middle Income
Sierra Leone SLE Low income
Solomon Islands SLB Lower Middle Income
Sri Lanka LKA Lower Middle Income
Sudan SDN Lower Middle Income
Swaziland SWZ Lower Middle Income
Tajikistan TJK Lower Middle Income
Togo TGO Low income
Uganda UGA Low income
Ukraine UKR Lower Middle Income
Uzbekistan UZB Lower Middle Income
Vietnam VNM Lower Middle Income
Yemen YEM Lower Middle Income
Zambia ZMB Lower Middle Income
Zimbabwe ZWE Low income
260 GROWTH AND CHANGE, MARCH 2018
TABLE A4. INSTRUMENT RELEVANCE TEST.
(1) (2)
Variables Neighbor Neighbor
Neighbor (IV) 0.850*** 1.144***
(0.102) (0.284)
Telephone fixed lines 0.135
(0.112)
GDP_cap 0.00329*** 0.00164
(0.000748) (0.00183)
Schooling 5.64*** 3.897
(1.28) (2.279)
Freedom of press 20.0165 0.0166
(0.0522) (0.0616)
Time to business 20.0101
(0.00887)
Export 20.0389
(0.0347)
Urban 0.526** 0.410
(0.213) (0.361)
Labor_female 20.381*
(0.224)
Unemployment 20.355**
(0.175)
Older than 64 23.03** 20.786
(1.35) (2.733)
Observations 353 433
Number of id 49 56
R-squared 0.703 0.671
Country FE YES YES
Robust standard errors in parentheses ***p < 0.01; **p < 0.05; *p < 0.1.
F test of excluded instruments Model 1 (full set of controls): F(1, 293) 5 69.70 Prob > F 5 0.0000.
Model 2 (limited set of controls): F(1, 363) 5 38.48 Prob > F 5 0.0000.
Angrist-Pischke multivariate F test of excluded instruments Model 1: F(1, 293) 5 69.70 Prob > F 5 0.0000.
Model 2: F(1, 363) 5 38.48 Prob > F 5 0.0000.
“BAD NEIGHBORHOOD” AND INTERNET ADOPTION 261
FIGURE A1. INTERNET ADOPTION IN 63 SAMPLE COUNTRIES FROM 2005 TO 2013.
Data Source: ITU.
262 GROWTH AND CHANGE, MARCH 2018
Copyright of Growth & Change is the property of Wiley-Blackwell and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
- l