Art 4

World Development Vol. 78, pp.
587609, 2016
0305-750X/ 2015 Elsevier Ltd. All rights reserved.
www.elsevier.com/locate/worlddev
http://dx.doi.org/10.1016/j.worlddev.2015.10.029
Has the Internet Fostered Inclusive Innovation

in the Developing World?
CAROLINE PAUNOV a and VALENTINA ROLLO b,*
a
OECD, France
b
International Trade Centre, Switzerland
Summary. The adoption of the Internet has been widespread across countries, making much more information available and thus
facilitating knowledge diusion among businesses to boost their innovation performance. However, dierences in rms capabilities to
use this newly available knowledge could create a new digital divide instead. Using 50,013 rm observations covering 117 developing
and emerging countries over the 200611 period, this paper tests for knowledge spillover eects from industries adoption of the Internet
on rms productivity and innovation performance. We test for heterogeneous spillover impacts on groups of rms that are commonly less
engaged in innovation and on rms with dierent productivity levels. Our specication regresses rm productivity and innovation
performance i.e., their investment in equipment and ownership of quality certicates and patents on industries use of the Internet.
Spillover eects are identied by controlling for rms own investment in Internet technology, industry and country-year xed eects
as well as extensive rm-level controls. Our results show that industries use of the Internet positively aects the average rms productivity
and its investment in equipment. We also identify modest impacts of industries use of the Internet on the likelihood that rms obtain
quality certicates and patents. On average, we nd that the returns to productivity are larger for rms that commonly engage less in
innovation, including single-plant establishments, non-exporters, and rms located in small agglomerations. However, results from
quantile regressions show that only the most productive rms reap productivity gains from Internet-enabled knowledge access. Firms with
productivity levels below the 50th percentile do not benet much. The spillover eects from industries adoption of the Internet identied in
our work justify public policies aimed at fostering industries use of the Internet. However, since we show that only rms with adequate
absorptive capabilities benet from the widespread Internet adoption, policy support should also focus on facilitating rms access to
networks and strengthening their capacities to use them.
2015 Elsevier Ltd. All rights reserved.
Key words Information and communication technology (ICT), knowledge spillovers, Internet, innovation, productivity, rm heterogeneities
1. INTRODUCTION
50,013 rm observations for 117 developing and emerging

countries for the 200611 period.
Our empirical methodology exploits information on industries adoption of the Internet as a tool for communicating
with suppliers and clients to identify Internet-enabled knowledge spillover eects. Our specication regresses rm productivity and innovation performance i.e., their investment in
equipment and ownership of quality certicates and patents
on industries use of the Internet, a comprehensive set of rm
controls as well as industry and country-year xed eects. We
control for rms own use of the Internet in all specications.
Our identication exploits within country-year dierences in
the adoption of the Internet across industries.
An industrys adoption of the Internet is unlikely to be
aected by an individual rms productivity and innovation
performance and, hence, the risk of reverse causality is low.
Notwithstanding, we cannot exclude a small risk of endogeneity which would arise if the most productive and innovative
rms use of the Internet caused other rms in the industry
The adoption of the Internet has been widespread across

countries (ITU, 2014). With the Internet, increasingly large
sets of knowledge and information (big data) can be more
easily diused to large groups of people. By allowing for wider
access to ideas, the Internet may boost innovation as innovation arises from new combinations of existing pieces of knowledge (Arthur, 2007). With the Internet, opportunities to
benet from knowledge created by others are possibly higher.
Such knowledge spillovers are critical for economic growth as
they generate increasing returns (Grossman & Helpman, 1991;
Krugman, 1991; Romer, 1986). In addition, the Internet can
support more inclusive innovation, i.e., the widening of the
often very small group of innovating rms in emerging and
developing economies (OECD, 2015; Paunov, 2013). The
Internet may help groups of rms that engage less in
innovation, by improving their access to knowledge. By
contrast, leading innovators often have access to quality
oine knowledge and consequently may have less to gain
from Internet-enabled knowledge transfers. However, less
innovative rms may lack capabilities to use newly available
knowledge for their business purposes (Cohen & Levinthal,
1989).
This paper provides evidence of knowledge spillover eects
from industries adoption of the Internet on rms productivity and innovation performance. It analyses whether the Internet as a conduit of knowledge spillovers has heterogeneous
impacts on groups of rms that dier with regard to their
innovation performance and productivity. This study provides
comprehensive evidence on these questions for a sample of
* The authors would like to thank Richard Baldwin, Nicolas Berman, Ana
Margarida Fernandes, Dominique Guellec, Eric J. Bartelsman and
participants of the 2014 ABCDE Conference, the Inter-American
Development Banks internal seminar series and the 7th Annual
Conference of the Academy of Innovation and Entrepreneurship (CAED)
for valuable comments. Valentina Rollo gratefully acknowledges support
from the SNF, Project Number PDFMP1_135148. The ndings expressed
in this paper are those of the authors and do not necessarily represent the
views of the OECD, the International Trade Center or their member
countries. Final revision accepted: October 3, 2015.
587
588
WORLD DEVELOPMENT
to adopt the Internet. Consequently, higher industry adoption

rates could be positively correlated with strong performance.
We address these endogeneity concerns as part of our robustness tests.
We use ordinary least squares regressions to study aggregate
eects on rm productivity and equipment investment rates as
well as logistic and probit regressions to analyze impacts on
rms ownership of quality certicates and patents. In addition, we apply quantile regressions to test whether benets
from industries adoption of the Internet dier across rms
of dierent productivity levels, which proxy for their capacities
to absorb new knowledge. We also use quantile regression to
ensure our results are not driven by non-normal errors and
outliers as might be the case for ordinary least squares regressions.
We nd that industries use of the Internet has positive
impacts on rms labor productivity and on their investments
in equipment. We also identify modest impacts on the likelihood that rms obtain quality certicates and patents. The
evidence is robust to various tests such as removing potential
outliers and using alternative sources of Internet-enabled
knowledge spillovers, including the Internet uptake by rms
within geographic locations and industries as well as at the
country rather than at the country-industry level.
Moreover, we show that, on average, the Internet adoption
of their industries benets more groups of rms that commonly engage less in innovation: rms that do not export,
rms that are not part of multi-plant establishments and rms
that operate in small agglomerations.
Quantile regression results show that the more productive
rms gain more from their industries intensive use of the
Internet. Firms with productivity levels below the 50th percentile do not benet much. Moreover, only the most productive non-exporting rms and single-plant establishments
benet from knowledge spillovers. Interestingly, while we do
not nd that rm size aects productivity gains of the average
rm, quantile regression results reveal larger payos for the
most productive small rms compared to larger rms.
Several policy implications arise from our analysis. First, the
existence of spillover eects from industries adoption of the
Internet, which do not depend on rms own investments, justify public policies aimed at fostering industries use of the
Internet. The potential returns from policy support of industries Internet adoption are high because, dierently from
other private sector development policies, benets arise even
for rms that commonly engage less in innovation and for
rms that face cumbersome business environment conditions
(Paunov & Rollo, 2015).
However, dierences in capabilities to use the knowledge
made available on the Internet could create a new digital
divide. Only rms with absorptive capabilities can benet
from business intelligence platforms, which give access to
knowledge relevant to their scientic and technological needs.
Therefore, facilitating rms access to such networks and
strengthening their capacities to use them deserve policy support. Human capital investments are core complementary
policies (Indjikian & Siegel, 2005).
This paper relates to the research on the contributions of
ICT to development. An ongoing debate focuses on adequate
infrastructure conditions in developing and emerging economies such as bandwidth capacity and aordable access prices.
Forbiddingly high prices to access the Internet, which may be
caused by technical or market imperfections, can reduce
uptake (Hilbert, 2010; Howard & Mazaheri, 2009). These
are pre-conditions for rms to use ICT for their business operations to support their productivity (Ref. Section 2(a)).
Moreover, our work relates to studies on the opportunities

for ICT to support very small rms and entrepreneurs from
disadvantaged socio-economic backgrounds. ICT have helped
smallholder farmers and sheries obtain market information
(including on price trends) as well as knowledge about production techniques (Jensen, 2007; Muto & Yamano, 2009; Ogutu,
Okello, & Otieno, 2014). Several studies have identied gains
from ICT-based services for disadvantaged producers (e.g.,
Aker & Mbiti, 2010; Donner, 2004, 2006; Donner &
Escobari, 2010; Duncombe & Heeks, 2002; Esselaar, Stork,
Ndiwalana, & Deen-Swarra, 2007; Kaushik & Singh, 2004).
However, improved access to knowledge has not always benetted these groups as they often lack capabilities to exploit
new knowledge (e.g., Tadesse & Bahiigwa, 2015). 1
In addition, our paper relates to the literature on knowledge
spillovers. Several studies have shown that gross social returns
to knowledge investments exceed private returns (Bloom,
Schankerman, & Van Reenen, 2013). Audretsch and
Feldman (2004) and Keller (2004) discuss research ndings
on the international and geographic dimensions of knowledge
spillovers. There is also evidence on barriers to knowledge spillovers, including from foreign direct investment (see Gorg &
Greenaway, 2004). Firms lack of absorptive capacity to
make use of newly available knowledge explains some of the
limitations (Girma, 2005; Kokko, 1994; Kokko, Tansini, &
Zejan, 1996). Geographical proximity also matters for spillovers (Ref. Section 2(c)).
Our paper makes several contributions to the literature. To
the best of our knowledge, this is the rst study to provide
comprehensive cross-country evidence of knowledge spillovers
of industries Internet adoption on rm productivity and innovation performance. Using data for the 200611 period allows
estimating global impacts at a point in time when the Internet
adoption gained some maturity. Data for earlier years may
underestimate impacts. Moreover, our study expands on the
previous analyses by testing whether groups of rms that commonly innovate less benet more from their industries Internet adoption. In addition, we apply quantile regression
techniques to explore whether average eects obtained from
conventional estimation techniques hide dierences in
impacts across rms of dierent productivity levels.
The remainder of the paper is organized as follows. Section 2
discusses the conceptual framework while Section 3 presents
the data we use for our analysis. Section 4 introduces the
empirical framework. Section 5 provides descriptive statistics
while Section 6 describes the results of the analysis. Section 7
concludes.
2. CONCEPTUAL FRAMEWORK
(a) ICT investments as driver of eciency improvements
Firms, industries, and countries that invested in ICT have
improved the eciency in which they transform inputs into outputs. Research, conducted at industry and rm levels, has consistently found that ICT investments positively relate to higher
productivity (e.g., Bartel, Ichniowski, & Shaw, 2007; Bloom,
Sadun, & Van Reenen, 2012; Jorgenson, 2001; Jorgenson &
Vu, 2005; Oliner & Sichel, 2000; Stiroh, 2002). 2 There is also
some evidence for rms in developing and emerging economies
(e.g., Commander, Harrison, & Menezes-Filho, 2011; ECLAC,
2011; Motohashi, 2008; Pohjola, 2001; UNCTAD, 2008; World
Bank, 2006). In addition, rms ICT investments also relate
positively to their innovation performance (see, for example,
Spezia, 2011, for an analysis of eight OECD countries).
HAS THE INTERNET FOSTERED INCLUSIVE INNOVATION IN THE DEVELOPING WORLD?
However, ICT do not automatically boost productivity. Evidence prior to the mid-1990s did not identify such positive
impacts (Brynjolfsson & Yang, 1996). Hence, Solows famous
quote (1987) you can see the computer age everywhere but in
the productivity statistics. Limited productivity gains also
reected the need for production process adjustments before
ICT had positive productivity impacts, including the need
for organizational changes and adequate human capital.
The complementarity between on the one hand ICT investment and on the other hand organizational change and management capacities is critical for ICT investments to boost
productivity (Black & Lynch, 2001, 2004; Bloom et al., 2012;
Bresnahan, Brynjolfsson, & Hitt, 2002; Bresnahan,
Greenstein, Brownstone, & Flamm, 1996). Dierences in organizational and managerial capabilities are among the reasons
why ICT contributed more to US than to Europes productivity (Cardona et al., 2013). Bloom et al. (2012) nd that for
rms operating in the United Kingdom, the productivity of
ICT capital has been signicantly higher in US-owned establishments compared to other rms.
(b) Knowledge spillovers and rms innovation performance
In addition to rms productivity improvements from their
own ICT investments, further productivity and innovation
performance gains may arise from industries adoption of
the Internet as a means of communication to improve the diffusion of knowledge. The Internets contribution is well illustrated by an analogy: drawing balls from an urn that holds
relevant knowledge. The Internet helps access a larger number
of balls from that urn. Improved communication among members of an industry facilitates learning about new technologies
and consequently accelerates the rate of innovation (e.g.,
Conley & Udry, 2010 and references therein).
Information from clients, suppliers, and competitors can
strengthen rms innovation performance in the following
ways: First, information on customer preferences helps identify market opportunities for new products and services.
Uncertainty about future market demands for new products
is a major obstacle for rms to invest in innovation
(Collard-Wexler, Asker, & De Loecker, 2011). User feedback
can also help rms develop new product and services; it is used
systematically to identify bugs in software codes and to create
improved software programs. 3 Second, developments of technology relevant to rms production processes determine the
technical feasibility of new products and processes. Hence,
better communication with suppliers to learn about new technological possibilities and discuss rms needs can support
innovation. Third, knowledge about competitors practices
allows rms to learn about alternative production techniques
and innovations. However, knowledge from competitors
might be less accessible because they have an interest in keeping information secret from each other (see e.g., Fernandes &
Paunov, 2012; Javorcik, 2004).
Aside from knowledge spillovers, there are other potential
sources of benet from industries adoption of the Internet
on rms productivity and innovation performance. Widespread adoption of the Internet allows rms to order online
from suppliers and deliver products more eciently to customers. The use of ICT can also improve the evidence-base
in rms decision-making (e.g., Brynjolfsson, Hitt, & Kim,
2011). Taken together these impacts are closely related to
rms own adoption of the Internet rather than to their industries adoption. Knowledge spillovers are likely to be more
important sources of productivity and innovation performance gains from industries Internet adoption.
589
(c) Knowledge spillovers and the Internet

Knowledge lends itself to spillovers since, once created, it
can be replicated and disseminated at virtually no cost, and
consequently benet more rms (Arrow, 1962). The Internet
has contributed to reducing dissemination costs further. However, there are barriers to Internet-enabled knowledge diusion: only codied knowledge can be transmitted while tacit
knowledge cannot (Leamer & Storper, 2001). This is why geographic proximity matters for knowledge diusion (Audretsch
& Feldman, 1996; Krugman, 1991). However, ICT have
reduced barriers for transmitting knowledge; for instance,
videoconference opportunities mimic face-to-face interactions
better than other ways of communicating across geographic
distances.
Potential benets from Internet-facilitated knowledge spillovers do not depend on the individual rms use of the Internet but on adoption by a critical mass of an industrys rms.
Even rms that do not use the Internet can benet from
Internet-enabled knowledge diusion within their industry
by other means, such as participation in business associations
and recruitment of sta from other rms. The question
whether the Internet facilitates knowledge spillovers is, therefore, distinct from asking how rms own adoption of ICTs
has beneted their performance.
(d) The Internet as a potential facilitator of inclusive innovation
More inclusive innovation i.e., expanding the group of innovators to groups that are traditionally not engaged in innovation, is an important issue for developing and emerging
economies. Small and young rms can be drivers of innovation as they often have greater agility to introduce novel ideas
(Acs & Audretsch, 1990). Yet, fewer opportunities for these
rms to access credit stie their contributions to innovation,
particularly in developing and emerging economies. Larger
incumbent businesses often have access to more funding
opportunities. Other obstacles are also less of a challenge for
large incumbent rms compared to small and young rms.
Consequently, a few incumbent rms are islands of excellence in a sea of smaller businesses of low productivity
(OECD, 2015). Such industrial structures are inecient as
aggregate productivity would be much higher if all rms were
as productive as the best performing ones (Hsieh & Klenow,
2009). Wage inequality is also higher as less productive rms
pay lower wages than the more productive rms (Mueller,
Ouimet, & Simintzi, 2015). Potential Internet-enabled knowledge diusion to improve the performance of less productive
rms is consequently critical.
The benets from the Internet adoption of their industries
may be larger for some rms than for others. Coming back
to the analogy of the urn introduced above, better knowledge
networks give rms access to all of the relevant knowledge
since the full number of available balls is xed. All else equal,
rms that are connected to rich (poor) oine knowledge networks may have fewer (stronger) productivity and innovation
performance gains from new online networks. Some groups of
rms dier in the quality of oine knowledge networks; those
with bad quality network connections commonly also engage
less in innovation. Dierences may arise for exporters and
foreign-owned rms, rms located in large (small) agglomerations, rms of dierent sizes as well as multi- and single plant
rms and informal businesses.
First, exporters and foreign-owned rms may have less to gain
from Internet-enabled knowledge spillovers because they access
foreign expertise, which is a critical source of advanced
590
WORLD DEVELOPMENT
technologies (Coe & Helpman, 1995; Fagerberg, 1994; Freeman

& Soete, 1997). National rms and non-exporters may consequently benet more from Internet-enabled knowledge transfer.
Second, rms located in large agglomerations often have
access to dense local networks, including frequent business
meetings with other companies located in the same cluster.
This is dierent for rms located in small agglomerations.
With the Internets ability to cross geographical distance better than previous communication technology (e.g.,
Cairncross, 1997; Forman & Van Zeebroeck, 2012;
Friedman, 2005), rms located in smaller agglomerations consequently stand to benet more. 4
Third, smaller-sized rms have fewer employees, lower revenues, and often smaller R&D investments in absolute terms.
Therefore, agglomeration benets from their own R&D
investments are lower as is the availability of internal sources
of knowledge (Cohen, 2010; Klepper & Simons, 2005). Smaller
rms may consequently benet more from Internet-enabled
knowledge spillovers than larger rms (cf. Acs, Audretsch, &
Feldman, 1994). Single-plant establishment may also have
more to gain from their industrys adoption of the Internet
than multi-plant establishments.
Fourth, informal businesses may also have larger benets
from Internet-enabled knowledge spillovers. These rms have
fewer resources to build knowledge networks and are often
excluded from formal businesses networks. There is evidence
to show ICT benet informal businesses (e.g., Jensen, 2007;
Muto & Yamano, 2009 and references provided in the introduction).
(e) Knowledge spillovers and absorptive capacities
Firms need the capacity to apply the knowledge they gain
access to. If absorptive capacities are weak, knowledge spillovers eects are much lower or absent (cf. Gorg &
Greenaway, 2004). Indigenous capacities are needed because
innovations developed elsewhere are often inappropriate in
specic rm contexts, unless incremental innovations to adjust
them are undertaken (Atkinson & Stiglitz, 1969). The empirical evidence conrms rms own capacities complements access
to knowledge (Hu, Jeerson, & Jinchang, 2005; Kokko, 1994;
Kokko, Tansini & Zejan, 1996). Thus, industries adoption of
the Internet may have heterogeneous eects on rms at dierent productivity levels: the most productive rms with stronger absorptive capacities may benet more from knowledge
spillovers than less productive rms that lack such capacities.
(f) Testable hypotheses
Based on the discussion, the following testable hypotheses
for our empirical analysis arise:
1. We expect the use of the Internet by industries as a tool
for communication to have positive impacts on rms
productivity and innovation performance.
2. We hypothesize that the Internet has heterogeneous
impacts on groups of rms that engage to dierent
extents in innovation: (i) exporting and non-exporting
rms, (ii) rms located in larger and smaller agglomerations, (iii) single- and multi-product rms and (iv) dierently sized rms. We also analyze (v) whether informal
businesses benets from their industries adoption of
the Internet.
3. We hypothesize that the impact of industries adoption
of the Internet on rms productivity and innovation
performance diers for rms of dierent productivity
levels to proxy for rms dierent levels of absorptive
capacities.
3. DATA
We use data from the second wave of the World Bank
Enterprise Surveys (WBES) for our empirical analysis. The
WBES collect information on a representative sample of formal rms in the countries non-agricultural sector; the selection of rms is done by stratied random sampling (Dethier,
Hirn, & Straub, 2011). The WBES have been widely used,
including in Almeida and Fernandes (2008), Beck,
Demirguc-Kunt, and Maksimovic (2008), Fisman and
Svensson (2007) and Paunov (2016). 5
Our analysis uses a cross-section of rms, information for
50,013 rm observations across 117 countries for the 2006
11 period. This sample is drawn from 65,285 rm observations
available, excluding observations with missing information on
labor productivity and industries use of the Internet. 6 Table 1
summarizes data coverage across world regions, industries,
rm size categories, years, and country income levels.
Table 1. Descriptive statistics
Number of
observations
Share in
total (%)
Region
Africa
Eastern Europe and Central Asia
Latin America and the Caribbean
Middle East
East Asia Pacic
South Asia
13,741
9,968
19,772
1,007
3,677
1,848
27.5
19.9
39.5
2.0
7.4
3.7
Industry
Food
Garments
Textiles and leather
Wood and furniture
Non-metallic and plastic materials
Metals, machinery and electronics
Chemicals and pharmaceuticals
Other manufacturing activities
Total manufacturing
6,326
3,987
2,567
689
2,337
3,738
2,387
6,921
28,952
12.7
8.0
5.1
1.4
4.7
7.5
4.8
13.8
57.9
Services (incl. construction)

Hotels and restaurants
Retail and wholesale trade
Construction and transportation
Other services
Total services
1,816
11,641
2,629
4,975
21,061
3.6
23.3
5.3
10.0
42.1
Size
Micro (110 employees)
Small (1150 employees)
Medium (51150 employees)
Large (more than 150 employees)
16,549
20,022
7,772
5,670
33.1
40.0
15.5
11.3
Year
2006
2007
2008
2009
2010
2011
12,280
8,261
2,382
14,057
11,182
1,851
24.6
16.5
4.8
28.1
22.4
3.7
Income level
High income
Upper-middle income
Lower-middle income
Low income
2,627
21,126
17,925
8,335
5.3
42.2
35.8
16.7
Full sample
50,013
Interestingly for our purposes, the WBES has information

on rms use of the Internet whether rms used email to
communicate with suppliers and customers rather than of
their investment in ICT, which says little about actual use.
The indicator informs about whether the Internet is used for
communication, which is critical for our study since it relates
to rms access to knowledge.
What is more, rms email use is positively correlated with
the intensity of their use of ICT in business operations.
Table 2 shows results using linear probability, logistic and
probit regression models for the sample of rms for which
we have information on whether they use the Internet to conduct R&D, purchase inputs from suppliers, deliver services to
clients, and market the rm on a website. 7 Results reported
in columns (1), (2), and (3) of Table 2 indicate that rm
email use is positively correlated with advanced uses of the
Internet i.e., those using the Internet in all these ways. By
contrast, as shown in columns (7), (8), and (9) of Table 2,
rm email use is negatively correlated with weak uses of
the Internet i.e., those using the Internet for none of the purposes described above. For median users the correlation is
positive but less strong than for advanced users (columns
(4), (5), and (6) of Table 2).
The WBES also has information on rms labor productivity and on innovation activities including their investments in
equipment as well as their ownership of quality certicates and
patents. (Information on patenting is unfortunately only available for a smaller number of observations because the variable
is not collected across all WBES surveys we combine in our
analysis.) The WBES also has ample information on rm characteristics, including on sales, employment, ownership type,
and export performance.
We also use the informal rm dataset, provided by the
WBES, which covers 1,557 rms for seven countries 8 in
2010. The informal business survey collects information on
rms uptake of mobile phones as well as on their sales and
their equipment investments.
4. EMPIRICAL FRAMEWORK
To study the impact of industries adoption of the Internet
on rms innovation and productivity performance, we use
the following estimation model:
591
Y ijct a b1 ICT jct b2 ICT ijct C X ijct kj kct eijct

1
where Yijct is a measure of rm is labor productivity or its
innovation performance, i.e., whether the rm owns a
quality certicate or patent and its level of equipment
investment. Coecient b1 is our variable of interest to
identify knowledge-spillover eects from the uptake of the
Internet: ICTjct is an indicator of industry js use of email
to communicate with clients and suppliers in country c in
year t. This measure is built for industries (by countryyear) with at least 10 observations. ICTijct is a measure
of rm is use of the Internet to control for direct impacts
of rms Internet adoption while kj and kct are respectively
a set of industry and country-year dummies. In consequence, our identication strategy exploits dierences in
industrys adoption of the Internet across countries while
controlling for characteristics specic to industries and
countries in any year.
We obtain a measure of country-year industry adoption,
identied across 15 dierent industries. Xijct is a vector of
rm-level control variables as described in Section 6. Acs
et al. (1994) and Haskel, Pereira, and Slaughter (2007) use similar empirical methodologies to study respectively the impacts
of industries R&D and FDI intensities. 9
Three possible challenges may be raised with regard to the
proposed empirical model: (i) endogeneity, that is, while ICT
might support innovation performance, more innovative rms
also rely more on ICT, (ii) omitted variable bias, i.e., there
might be other factors that aect the productivity-ICT adoption relationship and (iii) measurement error of the explanatory variable.
First, endogeneity is less of a challenge for our analysis
compared to an analysis of rms own investment in ICT
on their performance. It is unlikely that rms innovation
and productivity performance has a direct impact on their
industrys adoption of the Internet. To avoid potential
endogeneity concerns, we exclude rm is own use of the
Internet from the industry average we compute. Notwithstanding, we cannot exclude a small risk of endogeneity,
which could arise if the most productive and innovative
rms adoption of the Internet led other rms to use
the Internet. Higher industry adoption rates could then
be correlated with strong performance and explain possible
Table 2. Correlations between dierent intensities of rms ICT use and their email use
Dependent variables:
Advanced use of ICT
Median use of ICT
Low use of ICT
LPM
(1)
Logit
(2)
Probit
(3)
LPM
(4)
Logit
(5)
Probit
(6)
LPM
(7)
Logit
(8)
Probit
(9)
Email use
0.223***
0.690***
[0.166]
(0.104)
Yes
Yes
Yes
0.426***
[0.165]
(0.064)
Yes
Yes
Yes
0.388***
(0.013)
Yes
Yes
Yes
1.049***
[0.364]
(0.098)
Yes
Yes
Yes
0.165***
Firm-level controls
Industry xed eects
Country-year xed eects
1.920***
[0.405]
(0.199)
Yes
Yes
Yes
1.918***
[0.195]
(0.108)
Yes
Yes
Yes
1.147***
[0.216]
(0.064)
Yes
Yes
Yes
13,655
13,655
13,678
0.04
13,677
13,677
0.08
0.08
0.12
0.12
Observations
R2
Pseudo-R2
13,678
0.10
(0.024)
Yes
Yes
Yes
13,677
0.08
(0.023)
Yes
Yes
Yes
13,678
0.11
0.08
Note: Advanced, median, and low uses of ICT are dened depending on whether rms used the Internet to: (i) conduct R&D, (ii) purchase inputs from
suppliers, (iii) deliver services to clients and (iv) market the rm on a website. The table reports results from linear probability model regressions (LPM) in
columns (1), (4), and (7) and for logistic and probit regressions in columns (2), (5), and (8) and (3), (6), and (9), respectively. Robust standard errors are
shown in parentheses. For logistic and probit regressions, marginal eects are reported in brackets. Firm-level controls are the same as those of column (5)
of Panel A of Table 4. ***, **, and * indicate signicance at 1%, 5%, and 10% condence levels, respectively.
592
WORLD DEVELOPMENT
positive coecients on industry adoption rates. We

address endogeneity concerns as part of our robustness
tests.
Second, we address omitted variable biases by introducing
industry and country-year xed eects in addition to rmlevel controls, described in Appendix Table 11A. Countryyear xed eects allow isolating potential dierences across
countries in specic years. This includes government policies
with possible impacts on rms productivity and innovation
performance. Controlling for industry eects is also important
because certain industries are more technology-intensive than
others.
Third, measurement error is less of a concern than for
analyses that focus on interpreting rm-level evidence. Our
variable of interest is aggregated at the industry level and
consequently possible misreporting at rm-level is better
controlled for.
In order to test for possible heterogeneous eects across
rms we estimate the following modied model:
includes all explanatory variables as in (1), (2), and (3)

(Koenker & Bassett, 1978). Estimating h from 0 to 1 gives
the entire conditional distribution of Prodict, conditional on
zijct (Buchinsky, 1998). In other words, using quantile
regressions shows the impact of industries Internet
adoption at dierent levels of the conditional productivity
distribution, rather than at the conditional mean of the
dependent variable. Empirical applications of quantile
regression techniques include, for example, Yasar and
Morrison Paul (2007), Fattouh, Scaramozzino, and Harris
(2005) and Coad and Rao (2008). Quantile regressions also
allow for a robustness test of our main results as they are less
sensitive to outliers than standard regression models (Koenker
& Bassett, 1978). 10
5. DESCRIPTIVE STATISTICS
Many rms in developing and emerging countries have
adopted the Internet to support their operations over the
200611 period, but uptake rates diered between countries and rms. While 47% and 57% of the rms in
low-income and lower-middle-income economies communicated with clients and supplier by email, 84% and 93%
of the rms in upper-middle-income and high-income
economies did so. Moreover, small rms were less active
users than larger businesses: their uptake was of 44.5%
compared to 96.9% for businesses with more than 150
employees. Informal businesses were also active users of
mobile phones. Table 3 shows that 76.2% of the African
businesses in our sample used mobile phones in 200910
even though more than two thirds of these rms experienced power outages and more than one in four rms
did not have electricity.
The use of the Internet varied across dierent countries
industries. In the textiles industry, for instance, the share of
rms using the Internet for communication was of only
21% in Nigeria, 25% in Indonesia, and 33% in Pakistan while
the share was of 100% in Argentina, Costa Rica, and Peru.
In the retail and wholesale trade sector, the same shares
range from 20% for Uzbekistan and 30% for Angola to
almost complete adoption in Hungary (96%) and Estonia
(99%). Figure 1 shows substantial dispersion existed across
countries in industries adoption rates. 11 In the food, garment, and service industries i.e., retail and wholesale trade
and also hotels and restaurants several industries had
weakly adopted the Internet. By contrast, chemicals and
Y ijct a bType1 ICT jct Typeijct b1 ICT jct b2 ICT ijct

C X ijct kj kct eijct
Y ijct a bADV 1 ICT jct TypeADV ijct

bDIS1 ICT jct TypeDIS ijct b2 ICT ijct C X ijct
kj kct eijct
where Typeijct indicates rm characteristics (such as rm is

size) and TypeADVijct and TypeDISijct are dichotomous variables of rm characteristics (for instance, whether the rm is
an exporter, TypeADVijct, or not, TypeDISijct).
We use various estimation models in our analysis. To estimate impacts on the average rms labor productivity and
equipment investment we apply ordinary least squares regressions. We use logistic and probit estimation models to assess
the impacts of industry Internet adoption on the two binary
outcome variables: quality certicates and patents. Robust
standard errors clustered at country-industry-year level are
applied systematically to account for the fact that industry
adoption of the Internet is an aggregate variable (Moulton,
1990).
Moreover, we apply quantile regressions to assess whether
impacts dier based on rms level of labor productivity.
Quantile regressions can be expressed in the general form
Prodict = x0ict b + eict with Qh (Prodict/zijct) = z0ijct bh, where zijct
Table 3. Statistics on technology use of the informal sector

Overall
AFR
LAC
Firm nbr.
Firm nbr.
Firm nbr.
Use of cell-phone
No
Yes
1,026
1,495
40.7
59.3
295
943
23.8
76.2
674
489
58.0
42.1
Use of electricity
No
Yes
553
1,668
24.9
75.1
369
873
29.7
70.3
178
681
20.7
79.3
46.1
53.9
275
591
31.8
68.2
489
190
72.0
28.0
Experienced power outages

No
765
Yes
894
Note: The statistics are based on rm observations for 14 countries: Angola, Argentina, Botswana, Burkina Faso, Cameroon, Cape Verde, Democratic
Republic of Congo, Ivory Coast, Guatemala, Madagascar, Mali, Mauritius, Nepal, and Peru.
Percentage of firms using the Internet
Max
75th percentile
100
80
Median
60
25th percentile
40
20
Min
Figure 1. Dierences in industries adoption of the Internet across countries.

Note: The deciles for dierent industries are computed based on the
following number of country observations on the share of rms using the
Internet to communicate by email with clients and users: 110 for food, 50
for textiles, 71 for garments, 48 for chemicals and pharmaceuticals, 68 for
metals and machinery, 64 for non-metallic and plastic materials, 123
for retail and wholesale trade, 74 for hotels and restaurants and 81 for
construction and transportation. Statistics provided are obtained for the
50,013 rms included in our baseline sample.
pharmaceuticals industries had high adoption rates in most

countries.
6. RESULTS
(a) Baseline results: ICT-enabled spillovers on rm productivity
and innovation performance
First, we test whether the wider diusion of ICT leads to
knowledge spillovers that result in higher rm productivity
and improved innovation performance. Panel A of Table 4
shows regression results of Eqn. (1) for labor productivity: column (1) reports results for industries use of the Internet with
industry and country-year xed eects. We identify a positive
signicant eect. We progressively add rm-level controls:
rms employment and age (column 2), indicators of public
ownership and whether the establishments are part of multiplant establishments (column 3) and controls for whether the
rm has connections abroad (i.e., foreign-ownership and
exporter status) (column 4). We also add proxies for managerial quality and access to nance (column 5). Consistently with
the literature, we nd that these factors are positively correlated with rms productivity. Only public ownership is negatively correlated with rms productivity. We also control for
rms own use of the Internet to exclude direct eects on rms
(column 6). Our variable of interest, the industry-wide adoption of the Internet as a means of communication, remains
positive signicant but decreases as other factors, notably
rms own use of the Internet, are added to the specication.
Appendix Table 11B provides a correlation matrix of the independent variables; levels of correlation are modest and justify
the use of all dependent variables in our empirical estimations. 12
Panel B of Table 4 shows positive signicant eects on average rms investment in equipment (column 1). We also identify positive eects on rms ownership of quality certicates
(columns 2 and 3) and patents (columns 4 and 5). Both results
from probit and logistic estimation models are included. For
brevity we only report results from logistic estimation models
in subsequent tables. 13
Overall, our results provide evidence that industries adoption of the Internet facilitates positive spillover eects on
rms productivity and innovation performance. Results
593
conrm our rst empirical hypothesis. As for the magnitude

of estimated eects, all else equal, an increase by one standard
deviation in the intensity of a industries use of the Internet
improves rm labor productivity by an amount equivalent
to productivity increasing from the 50th to the 54th percentile
of the distribution. For equipment investment the increase is
from the 50th to the 55th percentile of the corresponding distribution. Impacts on rms ownership of quality certicates and
patents are modest. An increase by one standard deviation
would, all else equal, lead to an increase in formal intellectual
property rights ownership of 3% and 5% respectively.
(b) Robustness tests
This section presents robustness tests of our results. Findings for labor productivity are reported in Table 5. First, in
order to ensure that our variable of interest does not pick
up the eects of other industry characteristics, we include measures of national industries average employment, their age,
foreign ownership status, the volume of exporter activities,
an indicator of public ownership, the share of multi-plant
establishments, the average of years of managers experience,
and an indicator of credit access. Results, reported in column
(1) of Table 5, conrm our evidence is robust to including
these control variables. Unreported tests show our results also
hold if we include location xed eects and rms past productivity performance to account for a variety of possible omitted
factors.
Second, we check whether our results are robust to potential
outliers. We exclude the 5% most and least productive rms
for each country-year group. 14 Results, reported in column
(2) of Table 5, show outliers do not drive results. Results from
quantile regressions, reported in Section 6(d), provide additional evidence that outliers do not aect our ndings. A
related concern arises in those cases where few observations
are used to obtain averages of industry Internet adoption.
We, therefore, raise the threshold for averages to 30 observations. Results shown in column (3) of Table 5 conrm our
results are robust. The downside to raising the threshold is
that fewer observations can be included as part of the analysis.
This introduces a potential sample selection bias. This is why
we allow for a lower threshold for our main results. Third, our
main results focus on spillover eects within the rms industry, as product markets are sources of relevant knowledge for
rms productivity and innovation performance. The most
adequate sources of knowledge for rms may, however, be
not be rms own industries. In order to account for possible
alternative sources for knowledge spillovers we compute alternative measures, including separate measures of industries
adoption of the Internet for smaller and large rms. Smaller
rms may have more to gain from other rms of similar size
as processes adopted by large rms may be out of reach for
them. By contrast, large rms may have little to gain from
knowing about the practices adopted by smaller entities.
Results, reported in column (4) of Table 5, are positive significant and conrm our main ndings.
Moreover, we obtain separate measures of industries adoption of the Internet for dierent types of locations as in
Fisman and Svensson (2007). Firms in rural areas with very
few inhabitants may have more to gain from the practices of
other rms located in similar types of locations. 15 We nd a
positive signicant impact (column (5) of Table 5). In addition, unreported results using a measure of Internet adoption
for country-location-year level (as in Arnold, Mattoo, &
Narciso, 2008, and Dollar et al., 2006), are also positive significant.
594
WORLD DEVELOPMENT
Table 4. Baseline results
Panel A: Labor productivity

Dependent variable: Labor productivity
(1)
Industry Internet use
(2)
***
***
0.010
(0.001)
Firm-level controls
Employment
Age
(3)
(4)
0.007
(0.001)
0.007
(0.001)
0.006***
(0.001)
0.151***
(0.010)
0.083***
(0.011)
0.132***
(0.010)
0.078***
(0.011)
0.133*
(0.073)
0.333***
(0.022)
0.082***
(0.010)
0.087***
(0.011)
0.149**
(0.072)
0.280***
(0.022)
0.443***
(0.025)
0.258***
(0.022)
0.058***
(0.010)
0.078***
(0.011)
0.121*
(0.073)
0.285***
(0.022)
0.476***
(0.026)
0.241***
(0.022)
0.302***
(0.016)
0.026**
(0.011)
0.023**
(0.010)
0.075***
(0.011)
0.136*
(0.071)
0.254***
(0.022)
0.453***
(0.025)
0.191***
(0.021)
0.279***
(0.016)
0.027**
(0.011)
Foreign ownership
Exporter status
Credit access
Managerial expertise
***
Firm-level Internet use

Internet use
Industry xed eects
Observations
R2
Panel B: Innovation performance
(6)
0.008
(0.001)
Multi-plant rm
***
(5)
0.008
(0.001)
Public ownership
***
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
0.386***
(0.017)
Yes
Yes
56,169
0.79
55,121
0.80
52,839
0.80
52,146
0.81
50,107
0.81
50,013
0.81
Equipment investment
Quality certicates
Patents
OLS
(1)
Probit
(2)
Logistic
(3)
Probit
(4)
Logistic
(5)
0.009***
(0.003)
Yes
Yes
Yes
Yes
0.005**
[0.002]
(0.002)
Yes
Yes
Yes
Yes
0.007**
[0.002]
(0.003)
Yes
Yes
Yes
Yes
0.012**
Firm-level controls
Industry xed eects
0.003**
[0.001]
(0.001)
Yes
Yes
Yes
Yes
54,586
54,586
9,061
9,061
0.25
0.25
0.19
0.19
Observations
R2
Pseudo R2
33,080
0.45
(0.005)
Yes
Yes
Yes
Yes
Note: Panel A reports results from ordinary least squares regressions. Robust standard errors clustered at country-industry-year level are shown in
parentheses. For logistic and probit regressions, marginal eects are reported in brackets. Firm-level controls in Panel B are the same as those of column
(6) of Panel A. ***, **, and * indicate signicance at 1%, 5%, and 10% condence levels, respectively.
Finally, we compute a measure of Internet adoption at the

country level. This specication does not allow including
country xed eects. In order to ensure that other dierences
in countries business environment do not aect results, we
add a comprehensive set of country controls to our specications: GDP per capita, gross capital formation, net foreign
direct and portfolio investment, domestic credit to the private
sector (as % of GDP), school enrollment in primary and secondary education, literacy rates as well as world region dummies. Our results reported in column (6) of Table 5 are positive
signicant.
Fourth, we test whether exposure to better technology
improves impacts on rm productivity and innovation
performance. We interact our variable of interest with a

measure of industries use of imported technologies. As shown
in column (7) of Table 5, we nd that spillover returns from
the Internet are larger where the exposure to technology is
more important.
Finally, we check whether our results are dierent for rms
in the manufacturing and services sectors. As shown in column
(8) of Table 5, we nd positive signicant eects for both types
of rms but larger returns for services rms. This may be
because knowledge is more important for services than for
manufacturing rms. Alternatively, services rms may benet
more because better knowledge about customer preferences
may be more critical for them than for manufacturing rms.
Table 5. Robustness tests
Adding
context
controls
(1)
***
0.005
(0.001)
Controlling for outliers

Removing the
top and bottom
5%
(2)
***
0.005
(0.001)
Alternative aggregation
Setting the
threshold at
N P 30
(3)
Extensions
Firm size
Location
type
Country
level
(4)
(5)
(6)
Exposure
to
technology
(7)
0.006
(0.002)
0.007***
(0.001)
0.009***
(0.001)
Industry Internet use (location type)
0.244***
(0.018)
Country Internet use
0.006***
(0.001)
0.005***
(0.001)
Industry Internet use * high exposure to technology

Industry Internet use * low exposure to technology
Yes
Yes
Yes
Yes
No
Yes
Yes
Yes
Yes
No
Yes
Yes
Yes
Yes
No
Yes
Yes
Yes
Yes
No
Yes
Yes
Yes
Yes
No
Yes
Yes
Yes
No
Yes
0.00
Yes
Yes
Yes
Yes
No
0.005***
(0.001)
0.008***
(0.001)
0.01
Yes
Yes
Yes
Yes
No
49,790
0.81
44,959
0.87
44,115
0.81
44,476
0.82
41,442
0.82
9,777
0.83
50,013
0.81
50,013
0.81
Industry Internet Use * manufacturing

Industry Internet Use * services
Observations
R2
(8)
***
Industry Internet use (rm size)
P-value for the dierence in coecients

Firm-level controls
Industry xed eects
Country-level controls
Manufacturing
vs. services
Note: The table reports results from ordinary least squares regressions. Firm-level controls are the same as those of column (6) of Panel A of Table 4. Country controls included in regressions reported in
column (6) are country GDP per capita, the level of gross capital formation, national foreign direct and portfolio investment, domestic credit to private sector (as % of GDP), country school enrollment
in primary and secondary education as well as country literacy rates. Robust standard errors clustered at level of the Internet use variable are shown in parentheses. ***, **, and * indicate signicance at
1%, 5%, and 10% condence levels, respectively.

Controls
595
596
WORLD DEVELOPMENT
Robustness tests for our measures of innovation performance are reported in Appendix Table 12. As shown in Panel
A, robustness tests conrm ndings regarding rms investments in equipment. In this case, we do not nd impacts to
be dierent with regard to the exposure to technology (column
7), or across manufacturing and services rms (column 8).
With regard to quality certicates and patents, Panels B and
C of Appendix Table 12 show the evidence is less robust than
that on productivity and equipment investments. In the case of
quality certicates, manufacturing rms benet more than services rms. 16
(c) Testing for heterogeneous impacts of the Internet across
rms
We test our second hypothesis on dierential gains from
industry Internet adoption across dierent groups of rms.
We test for dierences in impacts across (i) exporters and
non-exporters, (ii) rms located in larger and smaller
agglomerations, (iii) single- and multi-product rms, and
(iv) smaller and larger rms. We also analyze whether (v)
informal businesses benet from their industries adoption
of the Internet.
Table 6A reports results for impacts on labor productivity.
We nd that there is more to be gained for non-exporters
(column 1 of Table 6A). Unreported results indicate that
national rms also benet more than foreign-owned rms.
Column (2) of Table 6A shows results where we distinguish
those rms located in countries capitals or in cities of more
than 1 million inhabitants from those located in smaller
agglomerations. We nd statistically signicant stronger
impacts on labor productivity for rms in small agglomerations, even if location eects are controlled for. Column (3)
of Table 6A shows that single-plant rms benet more compared to multi-plant rms, but the dierence is not statistically signicant. Finally, with regard to dierently sized
rms, reported in column (4) of Table 6A, we do not nd
evidence of heterogeneous eects when it comes to labor productivity.
The evidence for innovation indicators also points to differential eects, but is more mixed. We identify no eects
on exporters, as shown in columns (1), (5), and (9) of
Table 6B. The dierence in coecients is statistically significant for equipment investment and patenting. As for
location, while we nd that rms in small agglomerations
have larger returns to innovation performance, the
dierence in coecients is statistically signicant for rms
ownership of quality certicates. Results are shown in
columns (2), (6), and (10) of Table 6B. As reported in
columns (3), (7), and (11) of Table 6B we also nd
single-plant rms are more likely to obtain quality certicates and patents as a result of their industries adoption
of the Internet. We do not identify dierential impacts on
equipment investment rates. As for rm size dierences,
results reported in columns (4), (8), and (12) show no
signicant dierences in impacts except for results on rm
quality certicate adoption.
Finally, column (1) of Table 7 shows that informal businesses also benet from ICT-enabled knowledge spillovers in
terms of sales gains. We use the industrys adoption of mobile
phones as a proxy for the stronger use of ICT by informal
businesses. The evidence is maintained if we control for
Table 6A. Firm characteristics and benets from the Internets adoption: Labor productivity
Exporters
(1)
Industry Internet use * exporters
Industry Internet use * non-exporters
Firm location
(2)
Multi-plant rms
(3)
0.003*
(0.002)
0.006***
(0.001)
0.005***
(0.001)
0.008***
(0.001)
Industry Internet use * big agglomeration

Industry Internet use * small agglomeration
0.004***
(0.002)
0.006***
(0.001)
Industry Internet use * multi-plant rms

Industry Internet use * single-plant rms
Yes
Yes
Yes
Yes
0.00
Yes
Yes
Yes
Yes
0.01
Yes
Yes
Yes
Yes
0.08
0.006***
(0.001)
0.006***
(0.001)
Yes
Yes
Yes
Yes
0.45
50,013
0.81
44,706
0.82
50,013
0.81
50,013
0.81
Industry Internet use * bigger rms

Industry Internet use * smaller rms
Firm-level controls
Industry xed eects
P-value of the dierence in coecients
Observations
R2
Firm size
(4)
Note: The tables reports results from ordinary least squares regressions. Firm-level controls are the same as those of column (6) of Panel A of Table 4.
Robust standard errors clustered at country-industry-year level are shown in parentheses.***, **, and * indicate signicance at 1%, 5%, and 10%
condence levels, respectively.
Table 6B. Firm characteristics and benets from the Internets adoption: Innovation performance
Dependent variables
Exporters
Industry Internet
use * exporters
Industry Internet
use * non-exporters
OLS regressions
Logistic regressions
Firm Multi-plant rms Firm size Exporters Firm Multi-plant rms Firm size Exporters Firm Multi-plant rms Firm size
location
location
location
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
(11)
(12)
0.001
0.003
0.002
(0.005)
0.010***
[0.000]
(0.003)
0.006***
[0.000]
(0.006)
0.015***
(0.003)
[0.001]
(0.002)
[0.003]
(0.005)
Industry Internet
use * big agglomeration
Industry Internet
use * small agglomeration
0.008**
0.002
0.010**
(0.004)
0.011**
[0.000]
(0.002)
0.006**
[0.002]
(0.005)
0.014**
(0.004)
[0.001]
(0.003)
[0.003]
(0.006)
Industry Internet
use * multi-plant rms
Industry Internet
use * single-plant rms
0.009***
0.002
0.005
(0.004)
0.009***
[0.000]
(0.002)
0.007***
[0.001]
(0.006)
0.013**
(0.003)
[0.001]
(0.002)
[0.002]
(0.005)
Industry Internet
use * bigger rms
Industry Internet
use * smaller rms
Firm-level controls
Industry xed eects
P-value of the dierence in coecients
Observations
R2
Pseudo R2
Patents
0.009***
0.006***
0.011**
(0.003)
0.009***
[0.001]
(0.002)
0.004**
[0.002]
(0.005)
0.013**
Yes
Yes
Yes
Yes
0.00
Yes
Yes
Yes
Yes
0.31
Yes
Yes
Yes
Yes
0.05
[0.002]
(0.005)
Yes
Yes
Yes
Yes
0.31
Yes
Yes
Yes
Yes
0.01
Yes
Yes
Yes
Yes
0.44
Yes
Yes
Yes
Yes
0.89
(0.003)
Yes
Yes
Yes
Yes
0.93
31,281
0.45
27,612
0.44
31,281
0.45
31,281
0.45
Yes
Yes
Yes
Yes
0.18
Yes
Yes
Yes
Yes
0.08
Yes
Yes
Yes
Yes
0.00
[0.001]
(0.002)
Yes
Yes
Yes
Yes
0.01
54,586
49,048
54,586
54,586
9,061
9,061
9,061
9,061
0.25
0.25
0.25
0.25
0.19
0.19
0.19
0.19
597
Note: Firm-level controls are the same as those of column (6) of Panel A of Table 4. Robust standard errors clustered at country-industry-year level are shown in parentheses. For logistic regressions,
marginal eects are reported in brackets. ***, **, and * indicate signicance at 1%, 5%, and 10% condence levels, respectively.
(1)
Quality certicates
598
WORLD DEVELOPMENT
Table 7. The impact of industry cell phone use on the performance of informal businesses
Dependent variables
Sales
Industry cell phone use

Firm-level controls
Firm-level cell phone use
Industry xed eects
Machinery investment
(1)
(2)
(3)
(4)
(5)
(6)
0.010***
(0.003)
No
No
Yes
Yes
0.011***
(0.003)
Yes
No
Yes
Yes
0.010***
(0.003)
Yes
Yes
Yes
Yes
0.017**
(0.009)
No
No
Yes
Yes
0.016*
(0.009)
Yes
No
Yes
Yes
0.013
(0.009)
Yes
Yes
Yes
Yes
1,406
0.80
1,207
0.83
1,207
0.84
1,430
0.09
1,219
0.14
1,219
0.15
Observations
R2
Note: The table reports results from ordinary least squares regressions. Firm-level controls include employment size, their age, their ownership of bank
accounts and whether they had a loan. Robust standard errors are shown in parentheses. ***, **, and * indicate signicance at 1%, 5%, and 10%
condence levels, respectively.
rm-level control variables employment size, age, ownership

of bank accounts, and whether rms had a loan (column (2)).
Results are also robust to controlling for rms use of cell
phones (column (3). 17 However, the evidence of positive
impacts on informal rms machinery investments is weak
(columns (4)(6) of Table 7). Our estimated eects become
insignicant once we control for rms own use of cell phones
(column (6) of Table 7).
In conclusion, with regard to our second hypothesis, we nd
that the Internet provides larger opportunities for productivity
improvements of groups of rms that have limited access to
oine knowledge networks. This, however, does not hold
for dierently sized rms.
(d) Testing for the eects of absorptive capacities
We test our third hypothesis on the role of rms absorptive capacities for positive impacts from industry Internet
use on rm productivity and innovation performance. In
order to test whether the average rm eects identied hide
dierences in benets across rms of dierent productivity
levels, we conduct quantile regressions of Eqn. (1). Results,
which are reported in Figure 2, show that there are dierences
in the productivity gains from industries adoption of the

Internet. Gains are small for rms with productivity levels
below the 35th percentile but increase for more productive
rms, before leveling o at the 70th percentile. This evidence
supports our hypothesis that rms absorptive capacities
matter; the least productive rms have limited returns from
their industries Internet adoption.
In addition, we test whether average impacts across the rm
characteristics reported in Section 6(c) dier across the productivity distribution. With regard to exporter status, Figure 3
(a) shows highest dierential returns for non-exporting rms
at productivity levels above the median. While the gains for
non-exporters rise across the productivity distribution, benets for exporters are low for all producers, including the most
productive rms. In other words, the larger average impacts
on non-exporters performance identied in our previous analysis are driven by larger productivity gains for the most productive non-exporters. There is little dierence in (low)
benets among the least productive exporters and nonexporters.
Figure 3(b) plots the coecients of our variable of interest
for rms located in dierent agglomerations. The least productive rms reap minor productivity gains from industries Inter-
0.01
0.009
Coefficient
0.008
0.007
0.006
0.005
0.004
0.003
0.002
10
15
20
25
30
35
40
45
50
55
Quantiles
60
65
70
75
80
85
90
Figure 2. Impacts of the Internets adoption on productivity across the rm productivity distribution.
Note: The gure plots coecients from quantile regressions of the impact of the share of rms using email on labor productivity from the 10th to the 90th
quantile of the productivity distribution.
599
(a) By exporter status

0.010
Non-exporters
0.009
Exporters
0.008
Coefficient
0.007
0.006
0.005
0.004
0.003
0.002
0.001
10
15
20
25
30
35
40
45
50
55
Quantiles
60
65
70
75
80
85
90
(b) By agglomeration type

0.012
Firms in small agglomerations
0.010
Coefficient
Firms in big agglomerations

0.008
0.006
0.004
0.002
0.000
10
15
20
25
30
35
40
45
50
55
Quantiles
60
65
70
75
80
85
90
40
45
50
55
Quantiles
60
65
70
75
80
85
90
(c) By single- and multi-plant firms

0.009
Single-plant firms
0.008
Multi-plant firms
Coefficient
0.007
0.006
0.005
0.004
0.003
0.002
0.001
10
15
20
25
30
35
Figure 3. Dierent types of rms and the impact of the Internets adoption on productivity across the rm productivity distribution.
Note: The gures plots coecients from quantile regressions of the impact of industry in big and small agglomerations as in column (3) of Panel A of
Table 5 on labor productivity from the 10th to the 90th quantile of the distribution.
600
WORLD DEVELOPMENT
net adoption. By contrast, returns are larger for rms with

productivity levels above the median with larger gains for
rms located in smaller agglomerations. The gap between
rms located in bigger and smaller locations is largest for rms
in the median productivity range. Firms at higher levels of
productivity do not benet more from knowledge spillovers

than those at median productivity.
Figure 3(c) reports results for multi- and single-plant rms.
The evidence shows that the Internet provides limited returns
to the least productive single- and multi-plant rms. However,
Table 8. The eect of size on the impact of the Internets adoption across the productivity distribution
Quantile regression
Q1
(1)
Q2
(2)
Q3
(3)
Q4
(4)
Q5
(5)
Q6
(6)
Q7
(7)
Q8
(8)
Q9
(9)
Firm-level controls
Industry xed eects
0.000
(0.001)
0.003
(0.002)
Yes
Yes
Yes
Yes
0.001**
(0.000)
0.007***
(0.002)
Yes
Yes
Yes
Yes
0.002***
(0.000)
0.009***
(0.002)
Yes
Yes
Yes
Yes
0.002***
(0.000)
0.011***
(0.002)
Yes
Yes
Yes
Yes
0.002***
(0.000)
0.012***
(0.002)
Yes
Yes
Yes
Yes
0.003***
(0.000)
0.014***
(0.002)
Yes
Yes
Yes
Yes
0.003***
(0.000)
0.015***
(0.002)
Yes
Yes
Yes
Yes
0.003***
(0.001)
0.015***
(0.002)
Yes
Yes
Yes
Yes
0.003***
(0.001)
0.016***
(0.003)
Yes
Yes
Yes
Yes
Observations
R2
50,107
0.78
50,107
0.80
50,107
0.81
50,107
0.81
50,107
0.81
50,107
0.81
50,107
0.81
50,107
0.80
50,107
0.78
Industry Internet use * size

Note: Firm-level controls are the same as those of column (6) of Panel A of Table 4. Robust standard errors clustered at country-sector-year level are
shown in parentheses. ***, **, and * indicate signicance at 1%, 5%, and 10% condence levels, respectively.
Table 9. Impacts of productivity dierences on the Internet adoptions impact on rm innovation performance
Equipment
investment
Quality certicates
OLS regressions
(1)
Industry Internet use * below median
0.005
Industry Internet use * above median
(0.003)
0.007*
(2)
0.006
Industry Internet use * Q2
(0.003)
0.004
(0.003)
0.006*
(0.003)
0.008**
Observations
R2
Pseudo-R2
Yes
Yes
Yes
0.02
(4)
(0.004)
Yes
Yes
Yes
Yes
Yes
Yes
0.02
0.12
26,642
0.46
26,642
0.46
(5)
(6)
*
0.004
[0.000]
(0.002)
0.005**
[0.001]
(0.002)
(0.003)
Firm-level controls
Industry xed eects
P-value of dierence in coecients (below and above median)
P-value of dierence in coecients (between Q1 and Q4)
Logistic regressions
(3)
*
Patents
0.009
[0.002]
(0.005)
0.010**
[0.002]
(0.005)
0.004*
[0.001]
(0.002)
0.004
[0.000]
(0.002)
0.005*
[0.001]
(0.002)
0.006***
[0.001]
(0.002)
Yes
Yes
Yes
Yes
Yes
Yes
0.47
0.02
0.011**
[0.002]
(0.005)
0.008
[0.001]
(0.005)
0.010*
[0.002]
(0.005)
0.009*
[0.002]
(0.005)
Yes
Yes
Yes
0.40
41,720
41,720
7,087
7,087
0.26
0.26
0.18
0.18
Note: Firm-level controls are the same as those of column (6) of Panel A of Table 4. Robust standard errors clustered at country-industry-year level are
shown in parentheses. For logistic regressions, marginal eects are reported in brackets. ***, **, and * indicate signicance at 1%, 5%, and 10% condence
levels, respectively.
for the group of single-plant rms the benets rise for rms
with higher than median productivity. By contrast, for
multi-plant rms the rise is modest. As is the case of results
for exporters and non-exporters, we nd that the higher average gains reported in column (3) of Table 6A is driven by the
large returns for the most productive single-plant rms.
Table 8 shows quantile regression estimates, which include a
variable that interacts industry Internet adoption with rm
size. Dierently from average impacts reported in column (4)
of Table 6A, quantile regression results indicate that there
are dierences across smaller and larger rms: more productive smaller rms benet more than larger rms.
Finally, with respect to impacts on innovation, we nd,
as shown in Table 9, that the returns from industries
use of the Internet on equipment investment and ownership
of quality certicates are larger for rms with abovemedian productivity than for those with below-median
productivity. We nd no dierence with regard to rm
patenting.
7. CONCLUDING REMARKS
Using 50,013 rm observations for 117 countries over
the 200611 period, we provide evidence of a positive
601
impact of industries Internet use on rm performance.

These gains, which do not depend on rms own ICT
investments, justify public policies aimed at fostering industries use of the Internet. We also nd that the Internet
provide larger benets to rms located in smaller agglomerations, to single-plant establishments and to nonexporters. These rms commonly engage less in innovation
and consequently Internet-enabled knowledge spillovers
help increase the group of innovating rms. Positive eects
are the more so notable as they also arise where rms face
nancial constraints, frequent power outages, skills shortages, corruption, and cumbersome labor regulations
(Paunov & Rollo, 2015). However, complementary policies
aimed at building rms absorptive capacities matter if the
Internet is to support rms productivity and innovation
performance.
Several questions for future research arise, including how
increasingly sophisticated uses of the Internet inuence potential spillovers and returns to rm performance. Ensuring rm
surveys capture these uses is critical for research to better
assess impacts. Informal rm surveys are also important to
explore the potential of the Internet to help these businesses.
Questions related to innovation should be added as these rms
also engage in diverse non-technological innovation activities
(OECD, 2015).
NOTES
1. Ding, Levin, Stephan, and Winkler (2010) nd the Internet facilitated
the inclusion of women scientists and those working at non-elite
institutions in collaborative research. Agrawal and Goldfarb (2008)
show that the adoption of Bitnet, an early version of the Internet,
disproportionately beneted middle-tier universities collaboration with
leading universities.
2. Cardona, Kretschmer, and Strobel (2013) provide a comprehensive
review of the literature.
10. We analyze dierential impacts on other innovation variables by

interacting our variable of interest, industries adoption of the Internet,
with above or below median rm productivity at t 3 or, respectively, the
quartile of the distribution of productivity at t 3 the rm was part of
(Table 9).
11. As described in the notes of Figure 1 the number of country
observations diers across industries.
3. User involvement can also stimulate innovation if it leads to coinnovation with users (Bresnahan, Brownstone, & Flamm, 1996; Von
Hippel, 2005).
12. Unreported robustness tests show that results also hold if

only the variable of interest i.e., industrys use of the Internet and
rms own adoption of the Internet are included as part of the
analysis.
4. Forman, Goldfarb, and Greenstein (2014) conclude from their

analysis of Internet investment and patenting across US counties, that
the Internet has the potential to weaken the longstanding importance of
the geographic localization of innovative activity (p. 5).
13. All unreported results are available from the authors upon
request.
5. Dethier et al. (2011) provide a review of the WBES.
14. Our results are also robust to excluding outliers within industrycountry-year categories.
6. The routines used by the authors to clean the original dataset are
available upon request.
7. Only information on whether rms owned websites is also available
for the large sample of rms.
15. Unreported results for a measure of Internet adoption by locationtype, rm size, sector, country, and year are also positive signicant.
Aterido, Hallward-Driemeier, and Pages (2007) apply this approach in
their analysis.
8. These countries are Angola, Argentina, Botswana, the Democratic

Republic of Congo, Guatemala, Mali, and Peru.
16. We cannot report similar tests for patents, for which we have mainly
information on manufacturing rms.
9. S
eker (2012) and Dollar, Hallward-Driemeier, and Mengistae (2006)
apply this approach to analyze the impacts of business conditions on rm
performance.
17. We select a dierent set of control variables due to the dierent

nature of rms analyzed and the dierent variables contained in the
informal rm survey.
602
WORLD DEVELOPMENT
REFERENCES
Acs, Z. J., & Audretsch, D. B. (1990). Innovation and small rms. MIT
Press.
Acs, Z. J., Audretsch, D. B., & Feldman, M. P. (1994). R&D spillovers
and recipient rm size. The Review of Economics and Statistics, 76(2),
336340.
Agrawal, A., & Goldfarb, A. (2008). Restructuring research: Communication costs and the democratization of university innovation. The
American Economic Review, 98(4), 15781590.
Aker, J. C., & Mbiti, I. M. (2010). Mobile phones and economic
development in Africa. Journal of Economic Perspectives, 24, 207232.
Almeida, R., & Fernandes, A. M. (2008). Openness and technological
innovations in developing countries: Evidence from rm-level surveys.
The Journal of Development Studies, 44(5), 701727.
Arnold, J. M., Mattoo, A., & Narciso, G. (2008). Services inputs and rm
productivity in Sub-Saharan Africa: Evidence from rm-level data.
Journal of African Economies, 17, 578599.
Arrow, K. J. (1962). Economic welfare and the allocation of resources for
invention. In R. Nelson (Ed.), The rate and direction of inventive
activity (pp. 609626). Princeton University Press.
Arthur, W. B. (2007). The structure of invention. Research Policy, 36(2),
274287.
Aterido, R., Hallward-Driemeier, M., & Pages, C. (2007). Investment
climate and employment growth; The impact of access to nance,
corruption and regulations across rms. IZA Working Paper No. 3138.
Atkinson, A. B., & Stiglitz, J. E. (1969). A new view of technological
change. Economic Journal, 79(315), 573578.
Audretsch, D., & Feldman, M. (1996). R&D spillovers and the geography
of innovation and production. The American Economic Review,
630640.
Audretsch, D., & Feldman, M. (2004). Knowledge spillovers and the
geography of innovation. Handbook of Regional and Urban Economics,
4, 27132739.
Bartel, A., Ichniowski, C., & Shaw, K. (2007). How does information
technology aect productivity? Plant-level comparisons of product
innovation, process improvement, and worker skills. The Quarterly
Journal of Economics, 122(4), 17211758.
Beck, T., Demirguc-Kunt, A., & Maksimovic, V. (2008). Financing
patterns around the world: Are small rms dierent?. Journal of
Financial Economics, 89(3), 467487.
Black, S., & Lynch, L. (2001). How to compete: The impact of work-place
practices and information technology on productivity. The Review of
Economics and Statistics, 83(3), 434445.
Black, S. E., & Lynch, L. M. (2004). Whats driving the new economy?:
The benets of workplace innovation. The Economic Journal, 114(493),
97116.
Bloom, N., Sadun, R., & Van Reenen, J. (2012). Americans do it better:
US multinationals and the productivity miracle. The American
Economic Review, 102(1), 167201, February.
Bloom, N., Schankerman, M., & Van Reenen, J. (2013). Identifying
technology spillovers and product market rivalry. Econometrica, 81(4),
13471393.
Bresnahan, T. F., Brynjolfsson, E., & Hitt, L. M. (2002). Information
technology, workplace organization, and the demand for skilled labor:
Firm-level evidence. The Quarterly Journal of Economics, 117, 339376.
Bresnahan, T., Greenstein, S., Brownstone, D., & Flamm, K. (1996).
Technical progress and co-invention in computing and in the uses of
computers. Brookings Papers on Economic Activity. Microeconomics,
183.
Brynjolfsson, E., Hitt, L. M., & Kim, H. (2011). Strength in numbers: how
does data-driven decision-making aect rm performance? MIT, unpublished manuscript.
Brynjolfsson, E., & Yang, S. (1996). Information technology and
productivity: A review of the literature. Advances in Computers, 43,
179214.
Buchinsky, M. (1998). The dynamics of changes in the female wage
distribution in the USA: A quantile regression approach. Journal of
Applied Econometrics, 13(1), 130.
Cairncross, F. (1997). The death of distance. Harvard University Press.
Cardona, M., Kretschmer, T., & Strobel, T. (2013). ICT and productivity:
Conclusions from the empirical literature. Information Economics and
Policy, 25(2013), 109125.
Coad, A., & Rao, R. (2008). Innovation and rm growth in high-tech

sectors: A quantile regression approach. Research Policy, 37(4),
633648.
Coe, D. T., & Helpman, E. (1995). International R&D spillovers.
European Economic Review, 39(5), 859887.
Cohen, W. M. (2010). Fifty years of empirical studies of innovative
activity and performance. Handbook of the Economics of Innovation, 1,
129213.
Cohen, W., & Levinthal, D. (1989). Innovation and learning: The two
faces of R&D. Economic Journal, 99, 569596.
Collard-Wexler, A., Asker, J., & De Loecker, J. (2011). Productivity
volatility and the misallocation of resources in developing economies.
National Bureau of Economic Research.
Commander, S., Harrison, R., & Menezes-Filho, N. (2011). ICT and
productivity in developing countries: New rm-level evidence from
Brazil and India. The Review of Economics and Statistics, 93, 528541.
Conley, T. G., & Udry, C. R. (2010). Learning about a new technology:
Pineapple in Ghana. The American Economic Review, 3569.
Dethier, J.-J., Hirn, M., & Straub, S. (2011). Explaining enterprise
performance in developing countries with business climate survey data.
World Bank Research Observer, 26, 258309.
Ding, W. W., Levin, S. G., Stephan, P. E., & Winkler, A. E. (2010). The
impact of information technology on academic scientists productivity
and collaboration patterns. Management Science, 56(9), 14391461.
Dollar, D., Hallward-Driemeier, M., & Mengistae, T. (2006). Investment
climate and international integration. World Development, 34,
14981516.
Donner, J. (2004). Microentrepreneurs and mobiles: An exploration of the
uses of mobile phones by small business owners in Rwanda. Information Technologies and International Development, 2, 121.
Donner, J. (2006). The use of mobile phones by microentrepreneurs in
Kigali, Rwanda: Changes to social and business networks. Information
Technologies and International Development, 3, 319.
Donner, J., & Escobari, M. (2010). A review of evidence on mobile use by
micro and small enterprises in developing countries. Journal of
International Development, 22, 641658.
Duncombe, R., & Heeks, R. (2002). Enterprise across the digital divide:
Information systems and rural microenterprise in Botswana. Journal of
International Development, 14, 6174.
ECLAC. (2011), ICT in Latin America, United Nations, Santiago, Chile.
Esselaar, S., Stork, C., Ndiwalana, A., & Deen-Swarra, M. (2007). ICT
usage and its impact on protability of SMEs in 13 African countries.
Information Technologies and International Development, 4, 87100.
Fagerberg, J. (1994). Technology and international dierences in growth
rates. Journal of Economic Literature, 11471175.
Fattouh, B., Scaramozzino, P., & Harris, L. (2005). Capital structure in
South Korea: A quantile regression approach. Journal of Development
Economics, 76(1), 231250.
Fernandes, A., & Paunov, C. (2012). Foreign direct investment in services
and manufacturing productivity: Evidence for Chile. Journal of
Development Economics, 97(2), 305321.
Fisman, R., & Svensson, J. (2007). Are corruption and taxation really
harmful to growth? Firm level evidence. Journal of Development
Economics, 83, 6375.
Forman, C., Goldfarb, A., & Greenstein, S. (2014). Information technology
and the distribution of inventive activity. NBER working paper 20036.
Forman, C., & Van Zeebroeck, N. (2012). From wires to partners: How
the Internet has fostered R&D collaborations within rms. Management Science, 58(8), 15491568.
Freeman, C., & Soete, L. (Eds.) (1997). The economics of industrial
innovation. Psychology Press.
Friedman, T. L. (2005). The world is at: A brief history of the twenty-rst
century. Farrar, Straus and Giroux.
Girma, S. (2005). Absorptive capacity and productivity spillovers from
FDI: A threshold regression analysis. Oxford Bulletin of Economics and
Statistics, 67(3), 281306.
Gorg, H., & Greenaway, D. (2004). Much ado about nothing? Do
domestic rms really benet from foreign direct investment?. The
World Bank Research Observer, 19(2), 171197.
Grossman, G. M., & Helpman, E. (1991). Innovation and growth in the
world economy. Cambridge: MIT Press.

Haskel, J. E., Pereira, S. C., & Slaughter, M. J. (2007). Does inward
foreign direct investment boost the productivity of domestic rms?.
The Review of Economics and Statistics, 89(3), 482496.
Hilbert, M. (2010). When is cheap, cheap enough to bridge the digital
divide? Modeling income related structural challenges of technology
diusion in Latin America. World Development, 38(5), 756770.
Howard, P., & Mazaheri, N. (2009). Telecommunications reform, Internet
use and mobile phone adoption in the developing world. World
Development, 37(7), 11591169.
Hsieh, C.-T., & Klenow, P. J. (2009). Misallocation and manufacturing
TFP in China and India. The Quarterly Journal of Economics, 124(4),
14031448.
Hu, A. G., Jeerson, G. H., & Jinchang, Q. (2005). R&D and technology
transfer: Firm-level evidence from Chinese industry. The Review of
Economics and Statistics, 87(4), 780786.
Indjikian, R., & Siegel, D. (2005). The impact of investment in IT on
economic performance: Implications for developing countries. World
Development, 33(5), 681700.
ITU. (2014). 2014 facts and gures. Accessed at: <http://www.itu.int/en/
ITU-D/Statistics/Documents/facts/ICTFactsFigures2014-e.pdf>.
Javorcik, B. S. (2004). Does foreign direct investment increase the
productivity of domestic rms? In search of spillovers through
backward linkages. The American Economic Review, 94(3), 605627.
Jensen, R. (2007). The digital provide: Information (technology), market
performance, and welfare in the south Indian sheries sector. The
Quarterly Journal of Economics, 122, 879924.
Jorgenson, D. W. (2001). Information technology and the U.S. economy.
The American Economic Review, 91(1), 132.
Jorgenson, D. W., & Vu, K. (2005). Information technology and the world
economy. The Scandinavian Journal of Economics, 1074, 631650.
Kaushik, P. D., & Singh, N. (2004). Information technology and broadbased development: Preliminary lessons from North India. World
Development, 32(4), 591607.
Keller, W. (2004). International technology diusion. Journal of Economic
Literature, 42, 752782.
Klepper, S., & Simons, K. L. (2005). Industry shakeouts and technological
change. International Journal of Industrial Organization, 23(1), 2343.
Koenker, R., & Bassett, G. (1978). Regression quantiles. Econometrica:
journal of the Econometric Society, 3350.
Kokko, A. (1994). Technology, market characteristics, and spillovers.
Journal of Development Economics, 43(2), 279293.
Kokko, A., Tansini, R., & Zejan, M. (1996). Local technological
capability and productivity spillovers from FDI in the Uruguayan
manufacturing sector. The Journal of Development Studies, 32(4),
602611.
Krugman, P. (1991). Increasing returns and economic geography. The
Journal of Political Economy, 99(3), 483499.
Leamer, E. E., & Storper, M. (2001). The economic geography of the
Internet age. NBER Working Paper No. 8450.
Motohashi, K. (2008). IT, enterprise reform, and productivity in Chinese
manufacturing rms. Journal of Asian Economics, 19, 325333.
Moulton, B. R. (1990). An illustration of a pitfall in estimating the eects
of aggregate variables on micro units. The Review of Economics and
Statistics, 334338.
603
Mueller, H., Ouimet, P., & Simintzi, E. (2015). Wage inequality and rm
growth. NYU Stern School of Business, unpublished mimeo.
Muto, M., & Yamano, T. (2009). The impact of mobile phone coverage
expansion on market participation: Panel data evidence from Uganda.
World Development, 37(12), 18871896.
OECD (2015). Innovation policies for inclusive growth. Paris: OECD
Publishing.
Ogutu, S. O., Okello, J. J., & Otieno, D. J. (2014). Impact of information
and communication technology-based market information services on
smallholder farm input use and productivity: The case of Kenya.
World Development, 64, 311321.
Oliner, S. D., & Sichel, D. E. (2000). The resurgence of growth in the late
1990s: Is information technology the story?. Journal of Economic
Perspectives, 14, 322.
Paunov, C. (2013). Innovation and inclusive development: A discussion of
the main policy issues OECD Science, Technology and Industry
working paper no. 2013/1. OECD Publishing.
Paunov, C. (2016). Corruptions asymmetric impacts on rm innovation.
Journal of Development Economics, 118, 216231.
Paunov, C., & Rollo, V. (2015). Overcoming obstacles: The internets
contribution to rm development. World Bank Economic Review,
Papers & Proceedings, 29(Suppl. 1), S192S204.
Pohjola, M. (2001). Information technology, productivity, and economic
growth: International evidence and implications for economic development. Oxford University Press.
Romer, P. (1986). Increasing returns and long-run growth. Journal of
Political Economy, 94(5), 10021037.
S
eker, M. (2012). Importing, exporting, and innovation in developing
countries. Review of International Economics, 20, 299314.
Solow, R. (1987). Wed better watch out. New York Times Book Review,
July 12.
Spezia, V. (2011). Are ICT users more innovative? An analysis of ICTenabled innovation in OECD Firms. OECD Journal: Economic
Studies, 99119.
Stiroh, K. J. (2002). Information technology and the U.S. productivity
revival: What do the industry data say?. American Economic Review,
92, 15591576.
Tadesse, G., & Bahiigwa, G. (2015). Mobile phones and farmers
marketing decisions in Ethiopia. World Development, 68, 296307.
UNCTAD. (2008). Measuring the impact of ICT use in business: The case
of manufacturing in Thailand, New York and Geneva.
Von Hippel, E. (2005). Democratizing innovation. Boston: MIT Press.
World Bank. (2006). Information and communication for development:
Global trends and policies. Washington, DC.
Yasar, M., & Morrison Paul, C. J. (2007). International linkages and
productivity at the plant level: Foreign direct investment, exports,
imports and licensing. Journal of International Economics, 71(2),
373388.
APPENDIX
604
Table 10. Observations by country

Country
Percentage share
in total
Country
Observations
Percentage share
in total
Country
Observations
Percentage share
in total
199
659
116
1,790
262
291
114
120
193
146
90
215
681
252
502
1,077
1,171
310
265
320
96
135
120
1,702
1,774
517
91
408
462
561
165
134
289
836
884
91
232
47
108
0.40
1.32
0.23
3.58
0.52
0.58
0.23
0.24
0.39
0.29
0.18
0.43
1.36
0.50
1.00
2.15
2.34
0.62
0.53
0.64
0.19
0.27
0.24
3.40
3.55
1.03
0.18
0.82
0.92
1.12
0.33
0.27
0.58
1.67
1.77
0.18
0.46
0.09
0.22
The Gambia
Georgia
Ghana
Grenada
Guatemala
Guinea
Guinea-Bissau
Guyana
Honduras
Hungary
Indonesia
Iraq
Jamaica
Kazakhstan
Kenya
Kosovo
Kyrgyz Republic
Laos
Latvia
Lesotho
Liberia
Lithuania
Macedonia
Madagascar
Malawi
Mali
Mauritania
Mauritius
Mexico
Micronesia
Moldova
Mongolia
Montenegro
Mozambique
Namibia
Nepal
Nicaragua
Niger
Nigeria
153
243
475
129
858
192
133
136
595
248
1,122
707
225
400
636
200
154
271
211
88
111
209
292
336
83
654
214
275
2,454
35
327
336
60
440
307
328
633
85
1,865
0.31
0.49
0.95
0.26
1.72
0.38
0.27
0.27
1.19
0.50
2.24
1.41
0.45
0.80
1.27
0.40
0.31
0.54
0.42
0.18
0.22
0.42
0.58
0.67
0.17
1.31
0.43
0.55
4.91
0.07
0.65
0.67
0.12
0.88
0.61
0.66
1.27
0.17
3.73
Pakistan
Panama
Paraguay
Peru
Philippines
Poland
Romania
Russian Federation
Rwanda
Samoa
Senegal
Serbia
Sierra Leone
Slovak Republic
Slovenia
South Africa
Sri Lanka
St. Kitts and Nevis
St. Lucia
St. Vincent and the Grenadines
Suriname
Swaziland
Tajikistan
Tanzania
Timor-Leste
Togo
Tonga
Trinidad and Tobago
Turkey
Uganda
Ukraine
Uruguay
Uzbekistan
Vanuatu
Venezuela
Vietnam
Yemen
Zambia
Zimbabwe
843
587
719
1,464
944
260
304
717
183
35
479
327
126
165
243
895
462
117
130
129
152
259
247
388
82
102
107
308
835
515
544
907
320
81
158
953
300
434
547
1.69
1.17
1.44
2.93
1.89
0.52
0.61
1.43
0.37
0.07
0.96
0.65
0.25
0.33
0.49
1.79
0.92
0.23
0.26
0.26
0.30
0.52
0.49
0.78
0.16
0.20
0.21
0.62
1.67
1.03
1.09
1.81
0.64
0.16
0.32
1.91
0.60
0.87
1.09
WORLD DEVELOPMENT
Albania
Angola
Antigua and Barbuda
Argentina
Armenia
Azerbaijan
The Bahamas
Barbados
Belarus
Belize
Benin
Bhutan
Bolivia
Bosnia and Herzegovina
Botswana
Brazil
Bulgaria
Burkina Faso
Burundi
Cameroon
Cape Verde
Central African Republic
Chad
Chile
Colombia
Democratic Republic of the Congo
Republic of the Congo
Costa Rica
Ivory Coast
Croatia
Czech Republic
Dominica
Dominican Republic
Ecuador
El Salvador
Eritrea
Estonia
Fiji
Gabon
Observations
Table 11A. Description of variables used

Name
Dependent variables
Labor productivity
Patents
Firm-level controls
Employment
Age
Public ownership
Multi-plant rm
Foreign ownership
Exporter status
Credit access
Internet use
Industry
Variables used for robustness tests

High (low) exposure to technology
Interaction variables
Size
Big (small) agglomeration
Bigger (small) rms
Above (below) median of past productivity
Q1, Q2, Q3 and Q4 of past productivity
Mean
Std. dev.
Logarithm of the ratio of total annual sales over full time employment windsorized at the top and bottom 1% for any
country-year, reported in thousand USD
Logarithm of the sum of 1 and the ratio of total annual expenditure for purchases of equipment over full time
employment, reported in thousand USD
A dummy equal to one if the establishment has an internationally-recognized quality certication, such as ISO 9000
or 14000 certications
A dummy equal to one if the establishment has a registered patent and zero otherwise
18
2.17
2.1
1.08
0.21
0.39
Percentage share of plants using email to communicate with clients and suppliers in industry j of country c in year t.
Robustness tests include alternative measures for Internet (i) by industry, country-year and rm size, (ii) by industry,
country-year and location type, (iii) by industry, country-year and geographic location and (iv) by country-year
68.7
27.1
Logarithm of the plants full-time employment

Logarithm of the dierence between the year the survey was conducted and the year the plant was created
A dummy equal to one if the government or state own a share of 10% or more of the plant and zero otherwise
A dummy equal to one if the plant belonged to a rm that had at least one other plant and zero otherwise
A dummy equal to one if the share of foreign ownership is bigger or equal to 10% and zero otherwise
An indicator that is equal to one if the plant exports (direct or indirect)
Dummy variable is equal to one if the plant has a line of credit or loan from a nancial institution and zero otherwise
Logarithm of years of the managers experience
Dummy variable where the plant has its own website and uses email and zero otherwise
A variable indicating in which sector the plant is operating: (i) food, (ii) wood and furniture, (iii) textiles, (iv)
garments, (v) leather, (vi) non-metallic and plastic materials, (vii) chemicals and pharmaceuticals, (viii) electronics,
(ix) metals and machinery, x) auto and auto components, xi) other manufacturing, (xii) retail and wholesale trade,
(xiii) hotels and restaurants, (xiv) construction and transportation, (xv) other services
3.2 [25]
2.7 [15]
0.01
0.15
0.12
0.23
0.42
2.70 [15]
0.41
1.4
0.7
0.68
Indicator of whether the number of establishments that use foreign technology in the industry is above (below) the
average number of rms that use foreign technology, across industries
Establishments full time employment
Indicator of whether the plant is (not) located in the capital or a city of more (less) than 1 million inhabitants
Indicator of whether the plants full-time employment is above or below the median distribution
Indicator of whether the plants productivity at t 3 (windsorized at the top and bottom 1% for any country-year)
was above or below the median productivity distribution
Dummy variable indicating if the plants productivity at t 3 (windsorized at the top and bottom 1%) was in the rst
(Q1), second (Q2), third (Q3) or fourth (Q4) quartile of the productivity distribution
Variables used for the analysis of informal businesses

Sales
Logarithm of total sales of the establishment, windsorized at the top and bottom 1% for any country-year, reported in
thousand USD, value in parenthesis
Machinery investments
A dummy equal to one if the establishment invested in machinery and zero otherwise
Industry cell phone use
Share of establishments in a country sector who used cell phones for their operations
Employment
Logarithm of total employment of the business
Age
Logarithm of the dierence between the year the survey was conducted and the year the rm was created
0.5
1.41
Certicates
Description
0.23
51.18
21.52
0.46 [1.6]
0.68
1.98 [7.2]
0.90
(continued on next page)
605
606
Table 11A (continued)

Name
Bank account
Cell phone use
Loan
Sectors
Description
Mean
A dummy equal to one if the rm owned a bank account

A dummy equal to one if the rm used a cell phone for its operations and zero otherwise
Indicator of whether the rm had a bank loan or not
A variable indicating in which sector the rm is operating in (i) food, (ii) furniture, (iii) handicrafts, (iv) clothes and
shoes, (v) other manufacturing, (vi) construction, (vii) sales, and (viii) other services
0.15
0.59
0.12
Std. dev.
Note: Labor Productivity, Equipment Investment, and Sales are used in LCU, when included in regressions.

Employment
Age
Public ownership
Multi-plant rm
Foreign ownership
Exporter status
Credit access
Firm Internet use
Employment
Age
1.00
0.28
0.24
0.01
0.06
0.05
0.24
0.35
0.27
0.41
1.00
0.32
0.11
0.24
0.23
0.37
0.30
0.14
0.42
1.00
0.06
0.08
0.01
0.17
0.16
0.43
0.20
Public
ownership
Multi-plant
rm
Foreign
ownership
Exporter
status
Credit
access
Managerial
expertise
1.00
0.06
0.05
0.03
0.01
0.01
0.04
1.00
0.20
0.08
0.04
0.00
0.17
1.00
0.19
0.00
0.03
0.14
1.00
0.20
0.12
0.30
1.00
0.15
0.26
1.00
0.16
Note: Correlations reported are computed for the estimating sample described in Table 1.
Firm Internet use
1.00
WORLD DEVELOPMENT
Table 11B. Correlation matrix
607
Table 12. Robustness tests for innovation performance results

Dependent variable: Equipment investment
Controls
Adding context
controls
Panel A: Equipment investment

(1)
Removing the
top and
bottom 5%
(2)
Setting the
threshold
at N P 30
(3)
0.012***
(0.004)
0.008**
(0.003)
0.010**
(0.004)
Extensions
Firms
size
Location
type
Country
level
Exposure to
technology
Manufacturing
and services
(4)
(5)
(6)
(7)
(8)
0.015***
(0.002)
0.006*
Industry Internet use (location

type)
(0.003)
0.247***
(0.028)
0.009***
Industry Internet use * high

exposure to technology
(0.003)
0.009***
Industry Internet use * low

(0.003)
0.009***
Industry Internet
use * manufacturing
0.89
(0.003)
0.009**
(0.004)
0.96
Industry Internet use * services

P-value for the dierence in
coecients
Firm-level controls
Industry xed eects
Observations
R2
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
No
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
32,954
0.45
28,231
0.45
29,282
0.45
29,408
0.45
27,236
0.44
6,466
0.49
33,080
0.45
33,080
0.45
608
WORLD DEVELOPMENT
Table 12 (continued)
Panel B: Quality certificates

Dependent variable: Quality certicates
Controls
Adding context
controls
(1)
0.003
[0.000]
(0.002)

Removing the
top and
bottom 5%
(2)
***
0.007
[0.001]
(0.002)
Setting the
threshold
at N P 30
(3)
Extensions
Firms
size
Location
type
Country
level
Exposure to
technology
Manufacturing
and services
(4)
(5)
(6)
(7)
(8)
***
0.007
[0.001]
(0.003)
0.007***
[0.001]
(0.001)

type)
0.002
[0.000]
(0.002)
0.010
[0.001]
(0.010)
0.005**

[0.001]
(0.002)
0.003

[0.000]
(0.002)
0.008***
Industry Internet
use * manufacturing
0.04
[0.001]
(0.002)
0.002
[0.000]
(0.002)
0.00

coecients
Firm-level controls
Industry xed eects
Observations
Pseudo R2
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
No
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
54,344
0.25
44,280
0.23
47,945
0.25
48,527
0.26
45,417
0.25
10,711
0.25
54,586
0.25
54,586
0.25
609
Table 12 (continued)
Panel C: Patents
Dependent variable: Patents
Controls
Adding context
controls
(1)
0.009
[0.002]
(0.006)
Removing the
top and
bottom 5%
(2)
**
0.014
[0.003]
(0.006)
Setting the
threshold
at N P 30
(3)
Extensions
Firms
size
Location
type
Country
level
Exposure to
technology
Manufacturing
and services
(4)
(5)
(6)
(7)
(8)
0.007
[0.001]
(0.006)
0.013***
[0.002]
(0.003)

type)
0.006
[0.001]
(0.005)
0.080***
[0.015]
(0.013)
0.010**

[0.002]
(0.005)
0.008

[0.002]
(0.005)
0.012***
Industry Internet
use * manufacturing
0.11
[0.002]
(0.005)
0.004
[0.001]
(0.009)
0.05

coecients
Firm controls
Industry xed eects
Observations
Pseudo R2
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
No
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
9,019
0.19
7,406
0.18
8,002
0.18
8,019
0.19
8,335
0.19
3,614
0.17
9,061
0.19
9,061
0.19
Note: Panel A reports results from ordinary least squares regressions while Panels B and C report results from logistic regressions. Firm-level controls are
the same as those of column (6) of Panel A of Table 4. Robust standard errors clustered at the level of the Internet use variable are reported in parentheses.
For logistic regressions, marginal eects are reported in brackets. ***, **, and * indicate signicance at 1%, 5%, and 10% condence levels, respectively.
Available online at www.sciencedirect.com
ScienceDirect

Art 4

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Art 4

Uploaded by

Copyright:

Available Formats

World Development Vol. 78, pp.

Has the Internet Fostered Inclusive Innovation

50,013 rm observations for 117 developing and emerging

The adoption of the Internet has been widespread across

to adopt the Internet. Consequently, higher industry adoption

Moreover, our work relates to studies on the opportunities

HAS THE INTERNET FOSTERED INCLUSIVE INNOVATION IN THE DEVELOPING WORLD?

(c) Knowledge spillovers and the Internet

technologies (Coe & Helpman, 1995; Fagerberg, 1994; Freeman

Services (incl. construction)

HAS THE INTERNET FOSTERED INCLUSIVE INNOVATION IN THE DEVELOPING WORLD?

Interestingly for our purposes, the WBES has information

Y ijct a b1 ICT jct b2 ICT ijct C X ijct kj kct eijct

Median use of ICT

Low use of ICT

positive coecients on industry adoption rates. We

includes all explanatory variables as in (1), (2), and (3)

Y ijct a bType1 ICT jct Typeijct b1 ICT jct b2 ICT ijct

Y ijct a bADV 1 ICT jct TypeADV ijct

where Typeijct indicates rm characteristics (such as rm is

Table 3. Statistics on technology use of the informal sector

Experienced power outages

Percentage of firms using the Internet

HAS THE INTERNET FOSTERED INCLUSIVE INNOVATION IN THE DEVELOPING WORLD?

Figure 1. Dierences in industries adoption of the Internet across countries.

pharmaceuticals industries had high adoption rates in most

conrm our rst empirical hypothesis. As for the magnitude

Panel A: Labor productivity

Firm-level Internet use

Industry Internet use

Finally, we compute a measure of Internet adoption at the

performance. We interact our variable of interest with a

Table 5. Robustness tests

Controlling for outliers

Industry Internet use (location type)

Country Internet use

Industry Internet use * high exposure to technology

Industry Internet Use * manufacturing

Industry Internet use (rm size)

P-value for the dierence in coecients

HAS THE INTERNET FOSTERED INCLUSIVE INNOVATION IN THE DEVELOPING WORLD?

Dependent variable: Labor productivity

Industry Internet use * big agglomeration

Industry Internet use * multi-plant rms

Industry Internet use * bigger rms

HAS THE INTERNET FOSTERED INCLUSIVE INNOVATION IN THE DEVELOPING WORLD?

Industry cell phone use

rm-level control variables employment size, age, ownership

in the productivity gains from industries adoption of the

HAS THE INTERNET FOSTERED INCLUSIVE INNOVATION IN THE DEVELOPING WORLD?

(a) By exporter status

(b) By agglomeration type

Firms in big agglomerations

(c) By single- and multi-plant firms

net adoption. By contrast, returns are larger for rms with

productivity do not benet more from knowledge spillovers

Industry Internet use * size

Industry Internet use * above median

Industry Internet use * Q2

Industry Internet use * Q3

Industry Internet use * Q4

Industry Internet use * Q1

HAS THE INTERNET FOSTERED INCLUSIVE INNOVATION IN THE DEVELOPING WORLD?