WBWP Infra

Public Disclosure Authorized
Policy Research Working Paper 10350

The Impact of Infrastructure

on Development Outcomes
A Meta-Analysis
Vivien Foster
Nisan Gorgulu
Dhruv Jain
Stéphane Straub
Maria Vagliasindi
Infrastructure Chief Economist Office

March 2023
Policy Research Working Paper 10350
Abstract
This paper presents a meta-analysis of the infrastructure inequality and poverty, trade, education and health, popu-
research done over more than three decades, using a data- lation, and environmental aspects. The results allow for an
base of close to a thousand estimates from 201 papers update of the underlying parameters of interest, the “true”
conducted between 1983–2022, reporting outcome elas- underlying infrastructure elasticities, accounting for pub-
ticities. The analysis casts a wide net to include the transport, lication bias, as well as for heterogeneity stemming from
energy, and digital or information and communications both study design and context, with a particular focus on
technology sectors and the whole set of outcomes covered policy relevant subsectors and developing countries.
in the literature, including output, employment and wages,
This paper is a product of the Infrastructure Chief Economist Office. It is part of a larger effort by the World Bank to
provide open access to its research and make a contribution to development policy discussions around the world. Policy
Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The author may be contacted
at vfoster@worldbank.org, mvagliasindi@worldbank.org, ngorgulu@worldbank.org, stephane.straub@tse-fr.eu, and
jain.dhruv@tse-fr.eu.
The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development
issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the
names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those
of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and
its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent.
Produced by the Research Support Team

The Impact of Infrastructure on Development Outcomes:
A Meta-Analysis
Vivien Foster, Nisan Gorgulu, Dhruv Jain, Stéphane Straub, Maria Vagliasindi*
JEL: H54, O18, O47, Q40, R40
Keywords: Infrastructure, Transport, Energy, ICT, Meta-analysis.
*
Foster, Gorgulu, and Vagliasindi: World Bank. Jain and Straub: Toulouse School of Economics. We thank Sylvain Chabé-
Ferret for insightful suggestions, and Pedro Bom and David Tuesta for sharing their data. Stéphane Straub acknowledges
funding from ANR under grant ANR-17-EURE-0010 (Investissements d’Avenir Program).
1 Introduction
The academic debate on the impact of infrastructure on development outcomes has been ongoing
for several decades, at least since Aschauer published his seminal paper in 1989 (Aschauer, 1989).
It has fueled considerable interest in policy circles because of its relevance for addressing some of
the main development challenges of the time: by 2022, 755 million people around the world lack
access to electricity, 2 billion lack access to safely managed drinking water at home, 3.6 billion lack
access to safely managed sanitation in their home and half a billion still practice open defecation,
half of the world population does not use the internet, and around 1 billion live more than 2kms
away from an all-weather road. Of course, these gaps are mostly concentrated in the developing
world. For example, 16 of the top 20 electricity access deficit countries are located in Sub-Saharan
Africa. Lack of access to these critical infrastructures in turn often means that people are being
curtailed from opportunities to learn, receive good quality health care, access good job openings,
or develop a business.
While these facts may lead to the simple conclusion that more investment in infrastructure is
needed, they also raise complex questions. What is the most efficient way to extend access in each
sector? For electricity, should grid extension be prioritized, or is it more pressing to provide
reliable electricity supply to those that already have access? For transport, how do returns from
investment in rural roads compare with those directed towards upgrading critical interurban arteries
or alleviating congestion in cities? More broadly, given the limited resources available, do
investments in some sectors have higher social and economic returns than in others? Are these
trade-offs different depending on the context, the initial level of development for example?
To start providing answers to these questions, this paper performs a meta-analysis of the
infrastructure research done over the past three decades regarding the transport, energy, and digital
or information and communications technology (ICT) sectors, casting a wide net to include the
whole set of outcomes covered in the literature. Hundreds of papers and several literature reviews
have been published and their focus has evolved over time. The first generation of research mostly
analyzed growth and productivity effects, based on a production function framework and using
2
public capital data in a panel cross-section framework.1 More recently, with the increasing availability
of more granular information, including geospatial data, and the technology to process very large
databases, the focus has shifted to studies analyzing a wide array of outcomes, specific sectors, and
implementing robust micro-econometric identification strategies.
We built a database of close to a thousand estimates from 201 papers reporting infrastructure
elasticities. The papers included are the result of a systematic search through literature reviews and
publication databases and were produced between 1983 and 2022.
We start by providing funnel plots by categories of outcomes and sectors, as well as standard
FAT-PET-PEESE type estimates (Stanley and Doucouliagos, 2012), in order to establish both the
potential existence of publication bias and the specific “true” underlying effects of infrastructure in
each case. We then analyze in more detail the heterogeneity in the findings, related to both study
design and infrastructure sector and context characteristics.
Overall, the analysis leads to five main conclusions. First, the literature on infrastructure has
diversified hugely since the 1980s. From an almost exclusive initial focus on the output-elasticity
and the use of public capital data, it has extended to specific sectors, many different outcomes, and
a large variety of data sources and types. This has led to increasing heterogeneity in the literature
and has implications for the way we conduct this meta-analysis, leading us to focus mostly on
specific sub-samples.
Second, as is generally the case for many other economic topics, it is subject to significant
publication bias. Third, as the literature recognized the limitations and moved away from its initial
focus on public capital, the estimates it has produced have become relatively smaller. We report
average estimated elasticities that range between zero and 0.06 for most of the sector-outcome
categories we consider. However, we are careful in interpreting these magnitudes. As we discuss
in the concluding section, given the differences in the nature of infrastructure indicators used, it is
unclear whether lower elasticities necessarily entail smaller impacts or lower rates of return.
1
See Straub (2011) for a critical review, and Bom and Ligthart (2014) for a meta-analysis of that specific strand
of applications.
3
Fourth, we look at the average elasticities specifically for developing countries samples, finding
that they are larger than those for developed economies for the digital and transport sectors, but not
for cross-sectoral and energy studies. Finally, we discuss how these estimates could be leveraged to
produce marginal rates of return for specific sector-outcomes, highlighting the lack of relevant data
in this area, which appears to be the next priority in terms of research.
This study contributes to the literature on infrastructure by providing a systematic assessment

of the body of research reporting elasticities of infrastructure for a large set of outcomes and across
three sectors. In that sense, it aims at providing updated reference points regarding the quantitative
effect of infrastructure, useful for both researchers and practitioners engaged in infrastructure
projects.
The paper is structured as follows. Next, Section 2 describes the selection criteria used, the data
construction process, the variables included in the analysis, and provides general descriptive
statistics. Section 3 then presents the basic methodology used to analyze the data. Section 4
presents funnel plots as well as estimation results. Section 5 discusses the results and their
implications, and concludes.
2 Data
2.1 Studies included in the review: Search and selection criteria
This meta-analysis concentrates on the impact of infrastructure on a large array of economic and
social outcomes. Infrastructure is understood here as covering three main sectors:
• Transport, comprising roads, railroads, ports, and airports.
• Energy, covering production, transmission, and distribution of electricity to households and

firms (referred to as access from a demand point of view).
• Digital, also often referred to as information and communications technology, or ICT, and
including fixed and mobile phones, internet access and use, and backbone internet
development.
4
Broadly speaking, we include any study with a specification of the form:
Yit = α + δXit + θi + θt + εit (1)
where i is the aggregation level of the study, which we discuss below; t is the time unit; on the right-
hand side, the independent variable Xit is an indicator of infrastructure as defined above; and θi and
θt are observation-level, and time fixed effects, respectively. Note that not all studies always include
the full set of fixed effects. There might be several reasons for that. Some are cross-section analyses
and do not display within variation at the level of the units of observation; others cover several
time periods but do no report variation in the X variable of interest. Studies in this literature
commonly use instrumental variable techniques, such as two-stage least squares (2SLS) or
generalized method of moments (GMM), to address endogeneity. Others claim exogeneity based on
“natural experiment” type arguments. In all cases, we assess whether the study handles endogeneity
concerns in a credible way.
On the left-hand side, the dependent variable Yit is an economic or social outcome pertaining to
one of the following categories:
• Output, which includes indicators of production, income, expenditures, etc., in levels or

growth rates, as well as productivity. We categorize it as output micro when observations
are at the level of firms or households, and output macro when it is at a higher level of
aggregation (district, regions, or country).
• Labor market, including indicators of employment and wages.
• Inequality and poverty, such as Gini or poverty share indicators.
• Trade, including exports and imports, as well as extensive margin indicators of access to
external markets for example.
• Human capital, covering all indicators related to educational and health outcomes.
• Population, including levels as well as migration flows.
5
• Environmental variables, such as pollution, greenhouse gases (GHG) emissions, etc.
We started with a literature review of over 300 studies (reported in Foster et al, 2023). In order
to be able to perform a quantitative meta-analysis, we restrict the sample to studies that report
elasticities (where both the dependent and the independent variables are in log form) or semi-
elasticities (where the dependent variable is in log form, and the independent one of interest is a
dummy variable) that can be readily converted to elasticities.2
The starting point of the sample collection is Straub (2008), which covered 140 specifications
from 64 recent empirical papers. We then collected additional references from other meta-analysis
papers and literature reviews on transport, energy, and digital issues written since 2008.
These papers include Bom and Ligthart (2014) and García et al. (2017), which focus
specifically on production function studies. For transport-related infrastructure improvements, we
identified additional references from the surveys of Redding and Turner (2015), Redding and
Rossi- Hansberg (2017), and Berg et al. (2017). For energy, we collected papers from Burgess et
al. (2020) and Lee et al. (2020). Finally, for digital and ICT topics, we relied on Bertschek et al.
(2015), Greenstein (2020), and Vergara-Cobos and Malasquez (2023). We also added references
through searches for infrastructure-related keywords in Google, the World Bank, and the George
Washington University digital libraries, as well relevant recent working papers submitted to the
World Bank Infra4Dev 2022 Conference.
We include all the papers and reports that comply with the following basic requirements: they
are written in correct, understandable language, mostly in English with a few exceptions of pa-
pers written either in Spanish or French, they state the research question clearly, contain basic
information on the econometric specification allowing an assessment of its soundness, and report
estimates together with either standard errors or t-values, as well as number of observations used
in the estimations.
We report several estimates for each study. Although we aim at covering all relevant
specifications, we apply the following criteria to limit their number. When covering a given
2
Elasticities could be computed even when the independent variable is not a dummy variable, but the information needed to
do so, such as the sample average of the independent variable, is generally lacking.
6
econometric specification that reports several results successively including additional controls, we
pick only the last one including all the relevant controls. So for example, for a paper reporting five
OLS and five 2SLS estimations involving the same dependent and independent variables, in which
more controls or fixed effects are successively included, we report only the last OLS and the last
2SLS ones (usually the ones in the right-most columns on the respective tables). In the case of a
table reporting different ways of addressing endogeneity of the explanatory variable, say standard
2SLS and GMM, we report both estimates. We repeat this process for all relevant combinations
of right-hand and left-hand side variables but exclude robustness checks on sub-periods or sub-
samples of the main analysis. While this may seem quite parsimonious, we still end up with cases
where we include several dozen estimates from a single study.
Finally, we take care in harmonizing the reported estimates across studies. In particular, papers
addressing the same issue may report positive or negative effects depending on the way they code
the explanatory variables. For example, some papers use distance, as resulting from the building of
new roads, as an explanatory variable, in which case a negative coefficient means an improvement
of the outcome under study, while others directly use the reduction or the inverse of the distance
as regressor, finding positive effects in case of improvements. We adjust the signs accordingly.
In addition to the papers’ references and main estimates, we systematically code moderator
variables, grouped in two categories: study design characteristics, and infrastructure sector and
context characteristics.
Study design characteristics include the aggregation level (country-, region-, district-, or unit
specific-level data; firm or household data) and type of data used (monetary measures of
infrastructure such as public capital versus physical measures), the theoretical framework used as
reference, whether the data has a spatial dimension, and the type of econometric technique used,
with a specific focus on whether endogeneity is explicitly addressed and how.
Regarding the infrastructure sector and the context, we record the type of publication (peer-
reviewed or working paper) and its timing, the affiliation of the authors (academia, government,
international organizations, private sector), the sample covered (single or multi-country and
whether these are developing countries or not), the timeframe of the data, the category of the
dependent variables (see above) and of the independent variables (sectors and sub-sectors, and
7
specific types of variables used). A full list of the variables used in the dataset is in Appendix B.
2.2 Descriptive statistics
The final database includes 201 papers and 991 estimates. The full list of papers with the number
of estimates included in the analysis is in Table A.1 in Appendix A. Table 1 presents the
distribution of estimates, by sectors and types of outcomes. In terms of sector, the largest category
is transport, followed by studies using a cross-sectoral measure of infrastructure, energy, and
digital. It is also noteworthy that despite a trend towards diversification of outcomes, overall 875
of 991 estimates correspond to only three categories: micro and macro output, and labor market.
Transport studies are where this inclusion of new topics is more obvious, with 90 outcomes of 395
not belonging to the three main categories.
Tables 2 and 3 show the distribution of the signs of the estimates by sectors and types of
outcome. In both cases, a consistent share of between 80 and 90 percent of the reported estimates
are positive. In the case of outcomes, there are two notable exceptions: only 56 percent of estimates
in inequality / poverty related studies and 59 percent of estimates of population effects are reported
as positive. This must be considered in a context where the number of estimates included in our
data is however relatively small (16 and 34 respectively).
Finally, note that approximately half of all the estimates (496 of 991) correspond to developing
countries samples. This proportion is highest for energy (82 percent of the estimates) and transport
(59 percent), and much smaller for digital (30 percent) and cross-sectoral estimates (16 percent).
3 Methodology
This section presents the steps taken in the paper to quantitatively analyze the large number of
estimates gathered in our sample of studies. Specifically, we aim at deriving our own summary
estimates of the underlying “true” effect that can be inferred from existing studies, explain the
observed heterogeneity in the existing evidence, and also assess to the extent to which it may suffer
from publication bias.
These steps are applied to specific sub-samples, because the large numbers of sectors and types
of outcomes covered are likely to generate excessive heterogeneity in the complete sample, making
8
it more difficult to derive precise conclusions.3 We start by grouping estimates by sectors (cross-
sectoral, digital, energy, and transport) and by types of outcomes (macro output, micro output,
labor market, etc.).
When running estimations, we also look at specific sub-samples combining one sector and one
outcome, whenever enough observations are available, because combining homogeneous
dependent and independent variables categories is likely to reduce sample heterogeneity. There are
eight of these: two for the digital sector (micro and macro output), three for energy (micro and
macro output, and labor), and three for transport (micro and macro output, and labor). Finally, we
also zoom in on three specific sub-sectors: electrification with a specific focus on rural access,
roads and among these rural roads, and internet access, also including mobile phones, which in
many developing countries are the predominant form of internet access.
Assessing publication bias is critical for a correct appraisal of the literature. It has been shown
to be pervasive in most economic literatures.4 When publication bias is large, ignoring it may lead
to summary estimates of policy-relevant parameters that are way off their potential true value,
sometimes by a factor of two or three. We first address this using the basic FAT-PET-PEESE
approach proposed by Stanley and Doucouliagos (2012).5
The FAT-PET-PEESE approach consists of three steps. First, the Funnel asymmetry test (FAT)
provides an assessment of publication bias by looking at the relationship between the effect size
of the studies and their precision as measured by the standard error of the estimate. Under random
sampling theory, the two should be independent.
Visually, this translates into the expected symmetry of the funnel plot, where estimates are on
the horizontal axis, and the inverse of the standard error on the vertical axis. Under the hypothesis
of no publication bias, we expect the universe of points to be symmetrically distributed around a
3
We still present results for the whole sample at the beginning for completeness.
4
Publication bias stems both from the tendency of editors and referees to favor statistically significant results for
journal publication, and from the related “file-drawer” problem, as this generates incentives for authors to write up and
submit in priority these significant results. See Andrews and Kasy (2019) for a recent account.
5
It has been shown to perform best in reducing the bias, in the sense that it comes closer to the results of preregistered
replications (Kvarven et al., 2020).
9
vertical line proxying the “true” underlying effect.
Econometrically, the corresponding test relies on the following equation:
̂ 𝑖𝑠 = β0 + β1S Eis + εis

𝑄 (2)
where 𝑄̂𝑖𝑠 is the individual effect i from study s, and S Eis is its standard error. H0 : β1 = 0 is a test of
publication bias, and rejecting HO means that we cannot rule out the existence of such bias.
The variance of the effect, and therefore εis, is likely to vary across studies, generating obvious
heteroskedasticity. Stanley and Doucouliagos (2017) show that the equation is best estimated using
Unrestricted Weighted Least Squares. For simplicity, we follow Stanley and Doucouliagos (2019)
and estimate the unrestricted WLS by running the simple OLS regression derived from the one
above divided by the standard error, which becomes:
1
tis = β1 + β0 + ϵis (3)
S Eis
𝑄̂ ϵ
where tis = 𝑆𝐸𝑖𝑠 is the t-statistic of we assume that the error term ϵis = 𝑆𝐸𝑖𝑠 is now of constant variance.
𝑖𝑠 𝑖𝑠
Next, based on the same equation, the Precision-effect test (PET) indicates whether there is an
effect beyond potential publication bias. It consists of testing the hypothesis H0: β0 = 0. Rejecting
H0 means that there is a non-zero effect in the literature under consideration.
Stanley and Doucouliagos (2012) show however that the coefficient β0 is also likely to be biased
in the presence of publication bias. This leads to the third step, called precision-effect estimate with
standard error (PEESE), which uses the variance (S E2is ) instead of the standard error in (2), based
on the fact that a quadratic relationship appears to provide a better fit between effects and their
standard errors:
̂ 𝑖𝑠 = β0 + β1S Eis2 + εis

𝑄 (4)
which, after operating the same transformation as above, leads to the following PEESE estimating
10
equation:
1
tis = β0 + β1S Eis + ϵis (5)
S Eis
Here, rejecting HO : β0 = 0 again means that we accept the existence of a significant effect in the
literature of reference.
Note that several estimates from the same studies are sometimes included in our sample. To
account for the potential correlation of error, when estimating (3) and (5), we cluster standard
errors at the study level.
Finally, the last step is to explicitly account for the heterogeneity across studies, by including
in the estimations relevant moderator variables.
Formally, the error term in (2) can be expanded and written as νis + ϵis , where ϵis is sampling
noise, and νis is heterogeneity in treatment effect across studies due to differences in countries,
sector, period, techniques, etc. Given a set of moderator variables Mk capturing these differences,
we assume that there is no single true effect, but rather that it may vary depending on a number of
characteristics of the underlying sample in any given study or on the technical characteristics of the
study.6 We can write:
ν = Σ βk Mkis + υis
Equation 2 thus becomes:
̂ 𝑖𝑠 = β0 + Σ βk Mkis + β1S Eis + υis + ϵis

𝑄 (6)
Dividing again by the standard error, allows us to estimate by OLS the equivalent:
6
In practice, some of this variation may also be assigned to the publication bias, which may now vary across (sets of) studies.
Think, for example, of specific econometric techniques leading to less statistical significance overall and hence to lower
publication or circulation probability.
11
1 𝑀𝑘𝑖
ti = 𝛽1 + 𝛽0 𝑆𝐸 + ∑ 𝛽𝑘 + 𝑢𝑖 (7)
𝑖 𝑆𝐸𝑖
υis + ϵis
where uis = , which is again assumed to be of constant variance, and errors are clustered
S Eis
again at the study level. Note that moderator variables are allowed to vary within studies, although
this does not need to be systematic.7
4 Results
4.1 Funnel plots
In this section, we start by presenting funnel plots by sectors and by types of outcomes. The plots
include a vertical line at zero, as well as two solid curves, corresponding to the 5 percent
significance level on both sides.8 For visualization purpose, we exclude a few extreme values,
namely elasticities below -3 or above 3, and 1/standard errors above 200.9 Funnel plots are a good
way to visualize the distribution of estimates in the literature and the likelihood that it suffers from
publication bias.
Figure 1 first presents the funnel plot for the whole sample. As expected, less precise estimates,
i.e., those with larger standard errors at the bottom of the graph, are more dispersed. In addition,
it is immediately obvious that the graph is asymmetrical and there are more positive estimates,
especially among less precise ones. This is a first indication of publication bias. In addition, there
is evidence of bunching above the statistical significance lines, especially on the positive side,
indicating a high prevalence of studies with results just above the 5 percent significance level.
While the sheer number of studies makes this less visible in the whole sample, this sign of
publication bias is especially remarkable in the next plots by sub-samples.
Figures 2 to 5 present similar plots, breaking down the sample by sectors, i.e., whether the
treatment is one of the following: a cross-sectoral measure, digital, energy, and transport. In all
7
In practice, when adding moderators, we stick to equation 6, as it has been shown that the main effect is more
precisely estimated in that case.
8
For each value on the x-axis, this is given by y=1.96/x if x is positive, and y=-1.96/x if x is negative.
9
This leads us to exclude 84 observations in the whole sample: 28 for the digital sector, 15 for energy, and 35 for
transport. We formally address the robustness of the estimation results to outliers in Section 4.3 below.
12
cases, there is a larger mass of estimate in the right part of the graph, although to different degrees,
depending on the sub-sample of interest.
Finally, Figures 6 to 8 repeat the exercise for the sub-samples with a dependent variable
belonging to the output macro, output micro, and labor respectively. The conclusions are again
similar.
4.2 FAT-PET-PEESE estimations
Next, we estimate equations 3 and 5 to quantify both the potential publication bias and the
underlying effects of interest. We start again with the sector-level results, in Tables 4 and 5.
At that level of aggregation, our sector-level samples still suffer from significant heterogeneity
to the extent that they bundle together a number of different outcomes. Remember for example
that the transport literature in our data includes 34 estimates of micro output, 210 of macro output,
61 labor, 25 trade, and 33 population outcomes among others. To address this, we look at the
sector-outcome sub-samples for which enough observations are available: digital sector-micro and
macro output, energy-micro, -macro output, and -labor, and transport-micro, -macro output, and
macro output, energy-micro, -macro output, and -labor, and transport-micro, -macro output, and -
labor. The results for these sub-samples are in Tables 6 and 7 .
The significant coefficient of the constant in column 1 of Table 4 indicates that overall there is
significant publication bias in the infrastructure literature, with estimates skewed towards positive
values, consistently with the visual information from the funnel plot. When breaking down the
sample of estimates, publication bias also shows up in all the sub-samples focusing on the single
sectors (digital, energy, and transport), and is only rejected for the sample of cross-sectoral
estimates. Moving to Table 6 confirms the existence of a significant positive publication bias in
virtually all the sub-categories, namely again digital micro and macro output, energy micro and
macro output, and transport micro and macro output, with the exception of transport labor studies.
Next, the coefficient for the 1/se variable in Table 4 provides a first approximation for the true
effect once publication bias is controlled for. The results are broadly confirmed by the PEESE
estimates in Table 5, so we discuss these directly.
Significant underlying effects are estimated at 0.16 for the studies using cross-sectoral measures.
13
Strikingly, the estimated “true” elasticities appear to be much smaller once we move to single sector
studies: the reference values are significant and equal to 0.007 for energy, and 0.03 for transport,
and at 0.015 and not significantly different from zero for the digital sector.
Looking at Table 7 allows for a more precise distinction. For the digital sector, the estimated
underlying elasticity for micro output is equal to 0.04 and significant, while it remains
indistinguishable from zero for macro output. For energy, the effect appears to be significant only
for macro output, at 0.04, while it is small and very close to zero for micro output (0.006) and labor
outcomes (0.002), though significant at the 1 percent level for the latter. Finally, transport studies
looking at macro output have an underlying positive and significant effect of 0.05.
4.3 Robustness to outliers
In meta-analysis estimations, more precise estimates carry a larger weight. One issue is that
observations with extremely small standard errors may then have a disproportionate impact on the
outcome. In Figures 12 to 14, we present the results from a leave-one-out exercise. For each sub-
sample, the Figures plot the PEESE estimates obtained when dropping one observation at a time.
The values are presented in ascending order.
As can be seen, there are relatively few outliers overall. Focusing on sector-outcome categories,
there are three outliers in the case of transport (respectively for micro output, macro output, and
labor), which exclusion leads to larger estimated effects. An examination of the specific studies
involved reveals that in all cases, these observations have very small standard errors (i.e., values of
1/standard error of more than 200, going up to 7,000 in some cases). Of course, very small standard
errors may simply correspond to equally small coefficients. However, an examination of the data
shows in a majority of cases these do correspond to estimates that report extreme t-statistics of up
to 245. In light of this, we run robustness checks excluding observations with 1/standard error
larger than 200 and t-statistics larger than 4. Overall, this tags 41 observations, 6 cross-sectoral, 19
in digital, 4 in energy and 12 in transport.
Table 8 presents the results from this robustness exercise. We find slightly larger average
elasticities for digital micro output (0.06 instead of 0.04) and energy output macro (from 0.04 to
14
0.045), while the estimate for transport output macro is now quite smaller (0.02, down from 0.05).
4.4 Specific sub-sectors
Some sectors are of specific interest for development policies. In this section, we focus on three
of them, paying specific attention to the rural dimension when possible: electrification, roads, and
internet access. There are 110 estimates from electrification studies, of which 93 in a rural context,
270 estimates of the impact of roads, of which 38 in a rural context, and 77 estimates of the impact
of internet access, with a subset of 6 relating to backbone infrastructure. Finally, adding mobile
phones to internet access, on the grounds that they are the main access device to the internet in
developing countries, leads to a sample of 100 observations.
Funnel plots 9 to 11 present the distribution of estimates in these sub-samples, displaying again
a notable bias towards positive estimates.
Table 9 reports FAT-PET-PEESE estimates for all three sub-sectors. In each case, we start
by reporting the FAT-PET estimates to assess formally publication bias, followed by the PEESE
estimates. In the case of electrification and roads, for which a meaningful rural sub-sample exists,
we report the PEESE estimates adding an interaction for the rural sub-sample of observations. The
“true” effect for rural electrification (resp. roads) is then the sum of the main coefficient and that
of the interaction.
Columns 1, 4, and 7 confirm the positive publication bias for the three sub-sectors. In column
2, electrification estimates appear to generate an average effect very close to zero. However, in
column 3 we see that this in fact hides a positive effect of around 0.04 for electrification in general,
counterbalanced by a negative coefficient in rural areas, which brings the net effect for rural
electrification close to zero (0.006) but still significant. This is consistent with the conclusions
from some recent literature reviews, such as Lee et al. (2020). Note that of the 12 papers included,
11 analyze on-grid type of electrification, and only one looks at an off-grid, solar lamp program.
All the estimates correspond to access studies.
For roads, the average effect from the PEESE estimates in column 5 is around 0.025 but not
significant at conventional levels. In column 6, we see that it increases to around 0.07 for rural
15
roads.10 Finally, the PEESE estimate for internet access yields a significant effect, although is
very small, at 0.007 (0.004 when mobile phones are included).
4.5 Moderators
To understand in more detail the heterogeneity due to variation across studies in terms of study
design and context, we next present results based on equation (6).
We start by introducing the full list of moderators. As can be seen in Table 10, few of them
are significant and even fewer are consistent across sub-samples. Some results are noteworthy.
Regarding study design, in the whole sample both regional and more micro data, such as firm- and
household-level, command smaller estimates than the excluded category, which is country-level
data. However, the results are overturned for regional data in the cross-sectoral and digital studies
sample, and for micro data in the digital case. When infrastructure is measured with public capital
data, estimates are larger in the cross-sectoral case, but smaller for energy. The inclusion of fixed
effects appears to lead to larger estimates in the digital sample. Finally, and quite surprisingly,
studies that explicitly address endogeneity do not appear to yield significantly different results, and
the use of spatial data is generally not significant either.
Turning to the context, more recent studies do seem to boast larger estimates for energy.
Estimates corresponding to more recent samples are smaller, but only for cross-sectoral and energy
studies. Studies focusing on developing countries provide larger elasticities for energy, but smaller
ones for transport.
Overall, these results tend to confirm that the literature is characterized by a very large
heterogeneity justifying the choice to analyze sub-samples separately. In view of the difficulty of
finding a set of common moderators, in Table 11 we only include a variable indicating that the
study involves one of several developing countries, to elicit the relative effect in that context. The
last line reports an F-test of the null hypothesis that the net effect for developing countries sample
is equal to zero. As can be seen, it is rejected for the whole sample, and the digital and transport
sub-samples, with net elasticities of approximately 0.06 overall, 0.10 for digital studies, and 0.045
10
The sum of the two coefficients is significantly different from zero at the 10 percent level.
16
for transport studies. On the other hand, the test fails to reject that the net elasticities are equal to
zero for cross-sectoral estimates involving developing countries. Finally, the net effect for energy
appears to be negative and significant at 0.01.
5 Discussion and conclusion
A few key conclusions emerge from the results above. First, there is evidence of systematic
publication bias in this literature. Studies reporting positive results tend to be over-represented.
This comes out clearly in the visual representation in funnel plots and the analytical results.
Identifying this bias is important, as it allows us to isolate the residual “true” effect in different sub-
samples. Of course, this is not unusual in the economic literature, as many areas suffer from a similar
bias.11
Second, since the topic became of relevant policy concern in the 1980s, the literature has
diversified in many directions. It has moved from an initial focus on public capital measures of
infrastructure to other types of data, including physical measures and more granular micro- and
spatial data. As a result, the share of sector-specific studies has increased over time. On the
outcome side, a similar diversification has occurred, moving from an initial dominance of studies
looking at output or productivity effects, to more recent ones analyzing labor market effects, in-
equality and poverty, trade, education and health, or environmental aspects among others. While
these studies remain a minority, there is clearly a trend towards the multiplication of issues being
scrutinized under the infrastructure label.
This has important implications for any attempt to draw lessons from the literature, including
of course this meta-analysis. The diversification trend affecting both the dependent variables
(outcomes) and the independent ones (the sectoral aspects and the way they are measured) means
that this is an increasingly heterogeneous literature. This heterogeneity may translate across sub-
fields into differences in terms of publication bias, in terms of underlying elasticities, and in terms
of the effect of key moderators. A meta-analysis requires a certain degree of homogeneity of the
11
See, for example, Ioannidis et al. (2017).
17
studies it includes, in terms of methodology and the type of treatment considered, to yield
interpretable results. Because of this, we chose to make sub-fields the main focus of our analysis,
looking at specific sectors and, when possible, at combinations of sectors and specific outcomes.
When doing this, we indeed find evidence of the heterogeneity mentioned above.
Third, moving to the results, we find evidence that studies based on public capital measures
yield larger estimates. Cross-sectoral studies, which rely in more than 80 percent of the cases on
public capital measures, have an estimated average elasticity of 0.16, even after controlling for
publication bias. On the other hand, sector studies relying almost exclusively on physical, access
or usage measures yield elasticities that range between zero and 0.06 at most. This is in line with
several sector-level literature reviews, which find relatively small effects of the different types of
infrastructure. We also note that this does not seem to be due primarily to a change in methods or
more sophisticated identification strategies. The reliance on instrumental variables to address
endogeneity does not show up as significant when used as a moderator, and there is no clear time
trend in terms of the size of estimates. We remain cautious about interpreting this reduction in the
magnitude of estimates, given the change in the type of infrastructure indicators used in the latter
studies, from public capital to physical units or access and usage rates. Ultimately, we care about
the actual social rates of return of infrastructure, which, as discussed below, may not necessarily
be aligned with the elasticities reported here.
Which combinations of sectors and outcomes yield significant elasticities? When taking
sectors as a whole, the average elasticity come out at 0.01 for energy and 0.03 for transport but is
not significant for digital. When looking at sector sub-samples, digital-micro output has an
elasticity of 0.04, energy macro output of 0.04, while energy micro output and labor elasticities are
quite precisely estimated at zero, and transport only yields a significant elasticity for macro output,
at 0.05. Robustness checks excluding outliers based on extreme standard errors and t-statistics
values do not dramatically alter these results, although they yield a slightly larger estimate for
digital micro output at 0.06, and a smaller one for transport macro output at 0.02.
Alternatively, electrification appears to give an elasticity of 0.04, but this becomes very small
(slightly below 0.01) although still significant when looking at the subset of rural electrification
studies. Conversely, studies on the effect of roads give an elasticity of 0.07 for the subset of rural
roads. Finally, studies looking at the impact of internet and mobile phones roll-out give a very
18
small but significant elasticity. Again, whatever the sample we look at, the elasticities we estimate
appear smaller than those found in the early literature.12
Fourth, of specific interest to us is how these numbers would differ when focusing specifically
on studies based on developing countries samples. Here again, we find important heterogeneity.
For cross-sectoral studies, we fail to reject that the effect for developing countries is zero. On the
other hand, we do find larger and significant effects for the digital sector (around 0.10) and transport
studies (0.045). Finally, the elasticity for energy studies is positive at 0.02, but becomes negative
for the developing countries sample.
To fix ideas on the relative size of average effects in developing versus developed countries for
sectors and sector-outcome categories, respectively, in Figure 15, we plot the main effects from the
PEESE estimations in Tables 5 and 7 together with those from the sub-samples including only
developing countries observations. A few differences are visible. First, the meta-analysis estimates
are much larger for digital studies focusing on developing countries, with point estimates at 0.085
for micro output and 0.07 for macro output. Although these results are relatively precisely
estimated, it is worth noting that the corresponding samples are small (22 and 12, respectively).
Second, developing countries meta-estimates are similarly larger for transport macro output (0.06
versus 0.05) and for transport labor, although not significant in this last case. Finally, no
differences are found for energy studies.
Finally, there is an important issue regarding how to translate these elasticities into policy
recommendations. Deciding on which policies or sectors to prioritize, and possibly on potential
financing strategies, would require that we translate elasticities into specific rates of return (see
Henry and Gardner (2019)). This in turn implies the need for infrastructure capital stock figures.
Under a basic production function approach, the marginal rate of return for a specific type of
𝑌
infrastructure can be approximated by the following formula: mrr = γIn f . 𝐼𝑛𝑓, where γIn f is the
elasticity, Y is GDP, and Inf is the stock of infrastructure.

There are however several difficulties involved in finding suitable proxies for Inf. Regarding
public capital, different methods combine national account data, public budget data, and
12
To give a recent comparison, the meta-analysis by Bom and Ligthart (2014), which focuses on production function studies
using public capital as an input, concludes to an average output elasticity of 0.106.
19
information from the Private Participation in Infrastructure (PPI) database to estimate
infrastructure investments, which can then be used through inventory methods to compute
estimates of stocks. As discussed in Fay et al. (2019), each method generates both exclusion and
inclusion errors. In addition, public budget data is the only source allowing for sectoral breakdown.
This practically means that the only recent source available, namely IMF data on Gross Fixed
Capital Formation (GFCF), would be of little use here, not even considering the fact that it misses
most of the private investment that dominates the digital sector.
Alternatively, one could generate bottom-up estimates of the value of sectoral infrastructure
capital stocks by taking physical data on infrastructure stocks at the country level and applying
unit replacement cost values to convert these into replacement value capital stocks in monetary
units, as was done in Canning and Bennathan (2000) for the 1980s and 1990s. Such country-level
estimates would be needed to make meaningful conversions, as there are likely huge differences
in GDP-to-infrastructure ratios, with countries with very large infrastructure stocks, such as for
example China, boasting low values, possibly around two-thirds, while other very poor countries
with very infrastructure may have ratios of up to five. We intend to explore these important issues
in subsequent work.
20
References
Andrews, I. and Kasy, M. (2019). Identification of and correction for publication bias. American
Economic Review, 109(8):2766–94.
Aschauer, D. A. (1989). Is public expenditure productive? Journal of monetary economics,

23(2):177–200.
Berg, C. N., Deichmann, U., Liu, Y., and Selod, H. (2017). Transport policies and development.
Journal of Development Studies, 53(4):465–480.
Bertschek, I., Briglauer, W., H Ã Œschelrath, K., Kauf, B., and Niebel, T. (2015). The economic
impacts of broadband internet: A survey. Review of Network Economics, 14(4):201–227.
Bom, P. R. and Ligthart, J. E. (2014). What have we learned from three decades of research on the
productivity of public capital? Journal of economic surveys, 28(5):889–916.
Burgess, R., Greenstone, M., Ryan, N., and Sudarshan, A. (2020). The consequences of treating
electricity as a right. Journal of Economic Perspectives, 34(1):145–69.
Canning, D. and Bennathan, E. (2000). The social rate of return on infrastructure investments.
Policy research working paper series, 2390, The World Bank.
Fay, M., Han, S., Lee, H. I., Mastruzzi, M., and Cho, M. (2019). Hitting the trillion mark–a look
at how much countries are spending on infrastructure. Policy Research working paper; no. WPS
8730 Washington, D.C.: World Bank Group.
Foster,V., Gorgulu,N., Straub,S., Vagliasindi, M. (2023). The Impact of Infrastructure on
Development Outcomes: A Qualitative Review of Four Decades of Literature (English). Policy
Research working paper; no. WPS 10343 Washington, D.C.: World Bank Group.
García, V. A., Meseguer, J. A., Ortiz, L. P., and Tuesta, D. (2017). Infrastructure and economic
growth from a meta-analysis approach: Do all roads lead to rome. BBVA Research.
Greenstein, S. (2020). The basic economics of internet infrastructure. Journal of Economic

Perspectives, 34(2):192–214.
21
Henry, P. B. and Gardner, C. (2019). Global infrastructure: Potential, perils, and a framework for
distinction. NYU Stern School of Business.
Ioannidis, J. P., Stanley, T. D., and Doucouliagos, H. (2017). The power of bias in economics
research. Economic Journal, 127(605):F236–F265.
Kvarven, A., Strømland, E., and Johannesson, M. (2020). Comparing meta-analyses and
preregistered multiple-laboratory replication projects. Nature Human Behaviour, 4(4):423–
434.
Lee, K., Miguel, E., and Wolfram, C. (2020). Does household electrification supercharge economic
development? Journal of Economic Perspectives, 34(1):122–44.
Redding, S. J. and Rossi-Hansberg, E. (2017). Quantitative spatial economics. Annual Review of

Economics, 9:21–58.
Redding, S. J. and Turner, M. A. (2015). Transportation costs and the spatial organization of
economic activity. In Handbook of regional and urban economics, Vol. 5, pages 1339–1398.
Stanley, T. D. and Doucouliagos, H. (2012). Meta-regression analysis in economics and business.

routledge.
Stanley, T. D. and Doucouliagos, H. (2017). Neither fixed nor random: weighted least squares
meta-regression. Research synthesis methods, 8(1):19–42.
Stanley, T. D. and Doucouliagos, H. (2019). Practical significance, meta-analysis and the credibility
of economics.
Straub, S. (2008). Infrastructure and growth in developing countries: recent advances and research
challenges. World Bank policy research working paper, (4460).
Straub, S. (2011). Infrastructure and development: A critical appraisal of the macro-level literature.
The Journal of Development Studies, 47(5):683–708.
Vergara-Cobos, E. and Malásquez, E. A. (2023). Growth and Transformative Effects of ICTs
Adoption: A Survey. Policy Research Working Papers;10312. World Bank, Washington, DC.
22
Figures
Funnel plots by type of infrastructure
Figures 1 to 11 plot, for each estimate in the dataset in the corresponding sample indicated in the
graph’s title, the coefficient (horizontal axis) against the inverse of the standard error (vertical axis).
Figures 1 to 5 show samples based on the infrastructure sector covered in the respective estimations
(whole sample, public capital, cross-sectoral, digital, energy, transport), i.e., the sector of the
independent variable of interest. Figures 6 to 8 group estimates according to the outcome (macro
output, micro output, labor market), i.e., the dependent variable of interest. Finally, Figures 9 to 11
focus on the sub-samples of estimates for electrification, also distinguishing rural electrification,
roads, including rural roads, and internet access as well as backbone internet infrastructure. The
solid curves correspond to the 5 percent significance relationship, i.e., for each θ on the x-axis, the
hypothetical value of 1/se(θ) such that 1/se(θ) = 1.96/θ if θ > 0 and 1/se(θ) = 1.96/θ if θ < 0.
Elasticity coefficients larger than 3 or smaller than -3, and standard errors smaller than 1/200 are
excluded for readability.
Leave-one-out robustness plots
Figures 12 to 14 report graphically the results from a leave-one-out robustness exercise. They plot
PEESE estimates for sectors and sector-outcome categories respectively, after dropping one
observation at a time from the corresponding sample. Coefficients are arranged in ascending order
for easy visualization. The 95% confidence intervals are indicated by the dashed lines.
Level of development
Figure 15 presents plots of the estimates by sector (panel (a)), and sector-outcome (panel (b)), for
the full sample and the developing countries samples respectively.
23
Figure 1: Funnel plot - Whole sample
Source: Authors’ elaboration.
Figure 2: Funnel plot - Cross-sectoral studies
24
Figure 3: Funnel plot - Digital
Digital
200 150
1/Standard Error
100 50
0
-2 -1
Coefficient
Figure 4: Funnel plot - Energy
Energy
200 150
1/Standard Error
100 50
0
-1
Coefficient
25
Figure 5: Funnel plot - Transport
Transport
200 150
1/Standard Error
100 50
0
-1
Coefficient
Figure 6: Funnel plot - Output macro
Output macro
200 150
1/Standard Error
100 50
0
-1
Coefficient
26
Figure 7: Funnel plot - Output micro
Output micro
200 150
1/Standard Error
100 50
0
-2 -1
Coefficient
Figure 8: Funnel plot - Labor
Labor market
200 150
1/Standard Error
100 50
0
-1
Coefficient
27
Figure 9: Funnel plot - Electrification
Figure 10: Funnel plot - Roads
28
Figure 11: Funnel plot - Internet
29
Figure 12: Robustness: ‘Leave-one-out’ PEESE Estimates by Sectors
Notes: Figure plots PEESE estimates after dropping one observation at a time from the corresponding
sample. Coefficients are arranged in ascending order for easy visualization. 95% confidence intervals are
indicated by the dashed lines.
30
Figure 13: Robustness: ‘Leave-one-out’ PEESE Estimates by Sector-Outcomes
31
Figure 14: Robustness: ‘Leave-one-out’ PEESE Estimates by Sector-Outcomes
Notes: Figure plots PEESE estimates after dropping one observation at a time from the corresponding
sample. Coefficients are arranged in ascending order for easy visualization. 95% confidence intervals are
indicated by the dashed lines.
32
Figure 15: PEESE Estimates by level of development
((a)) Sector-level samples
((b)) Sector-outcome-level Samples
Notes: Figure plots PEESE estimates of the respective sub-samples. In the lower Panel, we do not show the
estimates of Energy-Labor and Transport-Labor categories for ease of visualization: both Energy-Labor
estimates are very close to zero (0.0021 and 0.0023) and have standard errors smaller than 0.0001;
Transport-labor estimate for developing countries is large and not significant (coeff.= 0.172, s.e.= 0.14).
95% confidence intervals indicated by the error lines.
33
Tables
Descriptive statistics
Tables 1 to 3 present descriptive statistics.
Estimations
Tables 4 to 11 present results of the FAT-PET-PEESE estimations for the different sub-samples
discussed in the text. In FAT-PET estimation tables, we present the Funnel asymmetry test (FAT)
and the Precision-effect test (PET) estimates. The FAT test fails to reject the existence of
publication bias if the coefficient of the constant is statistically different from zero. The PET test
then considers that there is a non-zero effect of the variable of interest if the coefficient of the
variable 1/se is statistically different from 0.
In the PEESE estimation tables, we present the precision-effect estimate with standard error
(PEESE) estimates. The PEESE test then considers that there is a non-zero effect of the variable of
interest if the coefficient of the variable 1/se is statistically different from 0. Robust standard errors,
clustered at the study level, are in parentheses (*** p<0.01, ** p<0.05, * p<0.1). Information specific
to each table, such as sample restrictions, are indicated in the notes at the bottom.
34
Table 1: Descriptive statistics: Sectors and outcomes
Cross-sectoral Digital Energy Transport Total

Output Micro 14 40 88 34 176
Output Macro 212 77 60 210 559
Labor Market 2 22 55 61 140
Inequality / Poverty 0 4 9 4 17
Trade 4 1 0 25 30
Human Capital 0 0 6 8 14
Population 0 1 0 33 34
Environment 0 0 1 9 10
Land Value 0 0 0 11 11
Total 232 145 219 395 991
Table 2: Descriptive statistics: Sign of estimates by sectors
Cross-sectoral Digital Energy Transport Total

Negative 39 19 20 59 137
Percentage 16,81% 13,10% 9,13% 14,94% 13,82%
Positive 193 126 199 336 854
Percentage 83,19% 86,90% 90,87% 85,06% 86,18%
Total 232 145 219 395 991
35
Table 3: Descriptive statistics: Sign of estimates by types of outcome
Micro output Macro output labor Ineq. /Poverty Trade Human capital Population Environment Land value Total
Negative 23 73 15 7 1 2 14 2 0 137
36
Percentage 13,07% 13,06% 10,71% 43,75% 3,33% 14,29% 41,18% 20,00% 0,00% 13,82%
Positive 153 486 125 9 29 12 20 8 11 854
Percentage 86,93% 86,94% 89,29% 56,25% 96,67% 85,71% 58,82% 80,00% 100,00% 86,18%
Total 176 559 140 16 30 14 34 10 11 991
Table 4: FAT-PET estimates
(1) (2) (3) (4) (5)

Whole sample Cross-sectoral Digital Energy Transport
1/se 0.0166 0.1624*** 0.0138 -0.0004 0.0230

(0.012) (0.021) (0.013) (0.001) (0.014)
Constant 4.0102*** -0.1170 3.3543** 3.1852*** 3.9796**
(0.957) (1.042) (1.288) (0.365) (1.624)
Observations 991 232 145 219 395

R-squared 0.090 0.566 0.142 0.000 0.061
Notes: FAT-PET estimates at the sector level. Robust standard errors, clustered at the study level, in
parentheses: *** p<0.01, ** p<0.05, * p<0.1
Table 5: PEESE estimates
(1) (2) (3) (4) (5)

1/se 0.0200 0.1613*** 0.0152 0.0065** 0.0311**

(0.013) (0.021) (0.013) (0.003) (0.013)
se 5.1975** 4.4989* 0.6238 8.0849** 7.9073***
(2.189) (2.262) (0.669) (3.794) (2.381)
Observations 991 232 145 219 395

R-squared 0.133 0.618 0.178 0.182 0.126
Notes: PEESE estimates at the sector level. Robust standard errors, clustered at the study level, in
parentheses: *** p<0.01, ** p<0.05, * p<0.1
Table 6: FAT-PET estimates by sector-outcome categories
(1) (2) (3) (4) (5) (6) (7) (8)

Digital Digital Energy Energy Energy Transport Transport Transport
Output micro Output macro Output micro Output macro Labor Output micro Output macro Labor
1/se 0.0337*** 0.0132 -0.0012 0.0069 0.0000 -0.0053 0.0433*** 0.0202

(0.001) (0.013) (0.002) (0.013) (0.000) (0.006) (0.012) (0.028)
Constant 1.3917*** 5.1878** 2.9683*** 4.1094*** 1.7525*** 3.3900*** 2.0648* 8.9880
(0.284) (2.190) (0.383) (1.215) (0.386) (0.576) (1.203) (5.568)
38
Observations 40 77 88 60 55 34 210 61
R-squared 0.380 0.126 0.003 0.009 0.000 0.025 0.481 0.020
Notes: FAT-PET estimates at the sector-outcome level. Data is grouped by sector-outcome categories, whenever the number of observations is
large enough. Robust standard errors, clustered at the study level, in parentheses: *** p<0.01, ** p<0.05, * p<0.1
Table 7: PEESE estimates by sector-outcome categories
(1) (2) (3) (4) (5) (6) (7) (8)

1/se 0.0429*** 0.0151 0.0062 0.0412*** 0.0021*** 0.0104 0.0474*** 0.0446

(0.003) (0.014) (0.005) (0.003) (0.000) (0.012) (0.010) (0.048)
39
se 0.1274 59.9045** 9.2492*** 20.5807*** 7.8994*** 4.4016** 7.9751* 13.4962**

(0.356) (24.600) (1.014) (0.836) (1.191) (1.946) (4.003) (6.213)
Observations 40 77 88 60 55 34 210 61
R-squared 0.496 0.181 0.243 0.707 0.163 0.142 0.571 0.100
Notes: PEESE estimates at the sector-outcome level. Data is grouped by sector-outcome categories, whenever the number of obser-
vations is large enough. Data is grouped by sector-outcome categories, whenever the number of observations is large enough. Robust
standard errors, clustered at the study level, in parentheses: *** p<0.01, ** p<0.05, * p<0.1
Table 8: PEESE estimates by sector-outcome categories - Robustness checks
(1) (2) (3) (4) (5) (6) (7) (8)

1/se 0.0557*** 0.0048 0.0057 0.0442*** 0.0015*** 0.0104 0.0203*** 0.0463

(0.018) (0.003) (0.006) (0.004) (0.000) (0.012) (0.008) (0.059)
40
se 0.1008 66.3753*** 9.2573*** 20.5037*** 7.9128*** 4.4016** 8.8426** 13.3924**

(0.338) (22.270) (1.015) (0.779) (1.190) (1.946) (4.187) (6.047)
Observations 38 62 87 58 54 34 200 59
R-squared 0.272 0.319 0.238 0.689 0.161 0.142 0.107 0.093
Notes: PEESE estimates at the sector-outcome level. Data is grouped by sector-outcome categories, whenever the number of observa-
tions is large enough. Observations with a value of 1/se larger than 200 and a t-statistic larger than 4 are excluded. Robust standard
errors, clustered at the study level, in parentheses: *** p<0.01, ** p<0.05, * p<0.1.
Table 9: Other outcomes
(1) (2) (3) (4) (5) (6) (7) (8) (9)

FAT-PET PEESE PEESE FAT-PET PEESE PEESE FAT-PET PEESE PEESE
Electrification Electrification Electrification Roads Roads Roads Internet Internet Internet+Mobile
1/se 0.0011 0.0071 0.0418*** 0.0133 0.0245 0.0241 0.0040*** 0.0065*** 0.0038***
(0.001) (0.004) (0.001) (0.015) (0.016) (0.016) (0.000) (0.001) (0.000)
1/se*rural -0.0358*** 0.0472
(0.003) (0.041)
se 3.2961 3.2002 9.1802*** 9.0080*** 4.2225** 4.8130**
41
(2.168) (2.097) (3.336) (3.221) (1.597) (1.871)

Constant 2.8504*** 5.3313** 2.5516***
(0.369) (2.195) (0.434)
Observations 110 110 110 270 270 270 77 77 100

R-squared 0.001 0.074 0.110 0.014 0.060 0.062 0.153 0.295 0.354
F-test main +rural = 0 3.94 (0.0726) 3.64 (0.0606)
Notes: FAT-PET and PEESE estimates for electrification, road, and internet outcomes. 1/se*rural indicates the interaction of the inverse of the standard
error with a dummy for observations corresponding to a rural sub-sample. Robust standard errors, clustered at the study level, in parentheses: ***
p<0.01, ** p<0.05, * p<0.1.
Table 10: PEESE estimates with moderators
(1) (2) (3) (4) (5)

Main effect 0.0326*** 0.1652** 0.0053 -0.0146 -0.0116
(0.011) (0.066) (0.012) (0.029) (0.022)
Published -0.0191 0.0992 0.0823*** 0.0192 -0.0509
(0.037) (0.081) (0.023) (0.020) (0.062)
Academic authors 0.0286 0.0008 -0.0870*** -0.0521*** 0.1872*
(0.031) (0.061) (0.024) (0.007) (0.103)
Govt authors 0.0535 0.2210*** 0.0634** -0.0026 0.1069
(0.034) (0.062) (0.026) (0.026) (0.070)
Intl org authors -0.0192 0.0645 -0.0409 -0.0438* 0.1201**
(0.023) (0.164) (0.047) (0.022) (0.057)
Public. year -0.0027 0.0081 -0.0024 0.0049*** 0.0056
(0.002) (0.008) (0.001) (0.001) (0.006)
Cross-country -0.0157 0.6575*** 0.0358 0.0526 -0.2286*
(0.036) (0.218) (0.023) (0.049) (0.127)
Developing country 0.0163 0.1401 -0.0035 0.1099** -0.2514*
(0.019) (0.107) (0.020) (0.053) (0.136)
Median sample time 0.0009 -0.0113** 0.0016 -0.0018** 0.0003
(0.001) (0.005) (0.001) (0.001) (0.001)
Regional data -0.0580*** 0.2266** 0.1571*** -0.1092*** -0.0030
(0.019) (0.092) (0.055) (0.032) (0.107)
District data -0.0338 0.0392 -0.0375 -0.0238 0.0412
(0.030) (0.106) (0.025) (0.028) (0.071)
Firm data -0.0149 0.2493 0.1079*** -0.0714*** -0.0253
(0.018) (0.219) (0.029) (0.022) (0.086)
HH data -0.0632 0.1641* 0.0096 0.0038
(0.048) (0.087) (0.026) (0.111)
Prod. Function -0.0221 0.0267 -0.0459*** 0.0745*** -0.1257
(0.024) (0.097) (0.017) (0.027) (0.107)
Spatial data 0.0402* 0.2903* -0.0666 -0.0099 -0.0253
(0.023) (0.162) (0.061) (0.036) (0.080)
Endogeneity -0.0084 0.0646 0.0111 -0.0058 0.0232
(0.016) (0.075) (0.010) (0.014) (0.026)
DiD -0.0302 0.4920** 0.0083 -0.0145 -0.0391
(0.028) (0.240) (0.026) (0.013) (0.077)
Fixed effects 0.0412 -0.0851 0.0809*** -0.0289 0.0673
(0.026) (0.081) (0.013) (0.022) (0.076)
Cointegration 0.0635** -0.1888 0.0706 0.0451 0.1006
(0.028) (0.133) (0.048) (0.034) (0.086)
Public capital 0.0936** 0.2797* -0.0347 -0.2043*** -0.0578
(0.047) (0.147) (0.026) (0.070) (0.060)
Public. bias 4.9496** 1.0483 0.5028 7.1738* 7.5006***
(2.114) (3.703) (0.646) (3.813) (2.635)
Observations 991 232 145 219 395

R-squared 0.439 0.599 0.859 0.359 0.307
Notes: PEESE estimates at the sector level, with the addition of moderator variables. All variables are demeaned,
except the constant (main effect) and the variance (public. bias). Robust standard errors, clustered at the study level,
in parentheses *** p<0.01, ** p<0.05, * p<0.1.
42
Table 11: PEESE estimates with restricted moderators
(1) (2) (3) (4) (5)

Main effect 0.0272*** 0.1189*** 0.0361*** 0.0230*** 0.0267***
(0.009) (0.017) (0.011) (0.002) (0.010)
Developing countries 0.0355* -0.1018*** 0.0639*** -0.0362*** 0.0189
(0.019) (0.032) (0.022) (0.004) (0.020)
Public. Bias 5.0561** 4.8866** 0.5674 7.9863** 7.9833***
(2.138) (2.329) (0.651) (3.781) (2.385)
Observations 991 232 145 219 395

R-squared 0.102 0.052 0.551 0.201 0.018
F-test main + developing = 0 5.62 (0.0188) 0.20 (0.657) 9.60 (0.003) 16.52 (0.0002) 2.97 (0.088)
Notes: PEESE estimates at the sector level, with the addition of the moderator variable “developing”, which takes
value 1 if the estimate relies on a developing country sample only. All variables are demeaned, except the constant
(main effect) and the variance (public. bias). Robust standard errors, clustered at the study level, in parentheses ***
p<0.01, ** p<0.05, * p<0.1.
43
APPENDIX
The Impact of Infrastructure on Development Outcomes: A
Meta-Analysis
i
A List of Papers
Table A.1: Description of the papers used in the meta-analysis
No Title Source Author(s) Year Sectors # Est.

1 Intranational trade costs, product scope and Journal of Development Eco- Abeberese and Chen 2022 Transport 3
productivity: Evidence from India’s Golden nomics
Quadrilateral project
2 On the road: Access to transportation infras- Journal of Development Eco- Abhijit Banerjee, Esther 2020 Transport 6
tructure and economic growth in China nomics Duflo and Nancy Qian
3 Farther on down the Road: Transport Costs, Review of Economic Studies Adam Storeygard 2016 Transport 3
Trade and Urban Growth in Sub-Saharan
Africa
4 The Impacts of rural electrification on labour IGC Working Paper Aevarsdottir et al 2017 Energy 4
supply, income, and health: Experimental
evidence with solar lamps in Tanzania
5 A normative analysis of public capital Applied Economics Ai and Cassou 1995 Cross-Sectoral 2
6 Knowledge and communications Regional Science and Urban Ake Andersson, 1990 Transport 10
infrastructure and regional economic change. Economics Christer Anderstig and
Bjorn Harsman
7 How does mobile phone coverage affect World Bank Economic Review Aker and Fafchamps 2015 Digital 1
farm-gate prices? Evidence from West
Africa
8 From Darkness to Light: The Effect of Economic Development and Akpandjar and Kitchens 2017 Energy 6
Electrification in Ghana, 2000 - 2010 Cultural Change
9 Empowering the powerless: Does access to Energy Economics Alex Acheampong, 2021 Energy 14
energy improve income inequality? Janet Dzator and
Muhammad Shahbaz
10 Transport Infrastructure and Welfare: An World Bank Policy Research Ali et al 2015 Transport 6
Application to Nigeria Working Paper
11 Infrastructure in Conflict-Prone and Fragile World Bank Policy Research Ali et al 2015 Transport 10
Environments Working Paper
12 Does public infrastructure affect regional Growth and Change Andrews and Swanson 1995 Cross-Sectoral 4
performance?
13 The effects of broadband internet expansion ILR Review Atasoy 2013 Digital 7
on labor market outcomes
14 The impact of mobile technology on eco- 30th European Conference of Bahia et al 2019 Digital 3
nomic growth: global insights from 2000- the International Telecommu-
2017 developments nications Society (ITS)
15 The Welfare Effects of Mobile Broadband World Bank Policy Research Bahia et al 2020 Digital 6
Internet: Evidence from Nigeria Paper
16 Does public capital affect private sector Economic Modelling Bajo-Rubio and 1993 Cross-Sectoral 1
performance? An analysis of Spanish case Sosvilla-Rivero
ii
Table A.1 – continued from previous page
17 Public capital stock and state productivity Empirical Economics Baltagi and Pinnoi 1995 Transport; Cross- 4
growth: Further evidence from an error com- Sectoral
ponents model
18 Household Electrification and Indoor Air Journal of Environmental Eco- Barron and Torero 2017 Energy 1
Pollution nomics and Management
19 Infrastructure and Long-Run Economic Empirical Economics Bazoumana Ouattara 2019 Cross-Sectoral 2
Growth: Evidence from Chinese Provinces and Yin-Fang Zhang
20 Trade Integration, Market Size, and Review of Economic Studies Benjamin Faber 2014 Transport 6
Industrialization: Evidence from China’s
National Trunk Highway System
21 Measuring the contribution of public Scandinavian Journal of Eco- Berndt and Hansson 1992 Cross-Sectoral 1
infrastructure capital in Sweden nomics
22 Mobile and more productive? Firm-level Telecommunications Policy Bertschek and Niebel 2016 Digital 4
evidence on the productivity effects of
mobile internet use
23 More bits–more bucks? Measuring the im- Information Economics and Bertschek et al 2013 Digital 2
pact of broadband internet on firm Policy
performance
24 How infrastructure and financial institutions Journal of Development Eco- Binswanger et al. 1993 Energy; Transport 6
affect agricultural output and investment in nomics
India
25 The Brasilia Experiment: The Heteroge- World Development Bird and Straub 2019 Transport 9
neous Impact of Road Access on Spatial
Development in Brazil
26 Roads and the Geography of Economic World Bank Policy Research Blankespoor et al 2017 Transport 15
Activities in Mexico Working Paper
27 The economic impact of ICT Report Bloom et al. 2010 Digital 5
28 Infrastructure Investment and Labor IMF Economic Review Brooks et al 2021 Transport 1
Monopsony Power
29 Road Network Upgrading and Overland Journal of African Economies Buys et al 2010 Transport 4
Trade Expansion in Sub-Saharan Africa
30 Inversioń en infraestructura pu´blica y crec- Ca´mara Chilena de la Con- Byron Idrovo Aguirre, 2012 Cross-Sectoral 4
imiento econo´mico, evidencia para Chile. struccioń
31 Contribution to productivity or pork barrel? Journal of Public Economics Cadot et al 2006 Transport 2
The two
faces of infrastructure investment
32 Contribution to productivity or pork barrel? Journal of Public Economics Cadot et al. 2006 Transport 2
The two faces of infrastructure investment
33 The Effects of Infrastructure Development World Bank Policy Research Calderon and Serven 2004 Energy; Transport; 3
on Growth and Income Distribution Working Paper Digital
34 Is infrastructure capital productive? A Journal of Applied Calderoń et al 2014 Digital; Transport 2
dyna mic heterogenous approach Econometrics
iii
35 Infrastructure’s Contribution to Aggregate World Bank Policy Research Canning 1999 Digital; Energy 2
Output Working Paper
36 The Social Rate of Return on Infrastructure World Bank Policy Research Canning and Bennathan 2000 Transport; Energy 2
Investments Working Paper
37 Rural Roads and Intermediated Trade: Mimeo Casaburi et al 2013 Transport 2
Regression Discontinuity Evidence from
Sierra Leone
38 Measuring (in a time of crisis) the impact of Applied Economics Castaldo et al 2018 Digital 1
broadband connections on economic growth:
an OECD panel analysis
39 Lighting up the last mile: The benefits and Working Paper Chakravorty et al 2016 Energy 8
costs of extending electricity to the rural
poor
40 Public Infrastructure and Economic Growth Working Paper, UMT INRA Charlot and Schmitt 2000 Cross-Sectoral 3
in France’s Regions ENESAD
41 Green infrastructure: The effects of urban American Economic Journal: Chen and Whalley 2012 Transport 3
rail transit on air quality Economic Policy
42 The Nexus between Infrastructure (Quantity International Review of Ap- Chengete Chakamera 2018 Cross-Sectoral 4
and Quality) and Economic Growth in Sub plied Economics and Paul Alagidede
Saharan Africa
43 Infrastructure and growth in the European European Planning Studies Chiara Del Bo and 2012 Digital; Transport; 25
Massimo - Florio Cross-Sectoral
Union: an empirical analysis at the regional
level in a spatial framework
44 Roads, exports and employment: Evidence Journal of Development Christian Volpe 2017 Transport 5
Economics Martincus Jeronimo
from a developing country Carballo
and Ana Cusolito
45 ICT services and small businesses’ Information Economics and Colombo et al 2013 Digital 2
productivity gains: An analysis of the Policy
adoption of broadband Internet technology
46 Electricity shortages and manufacturing Energy Policy Corbett Grainger and 2019 Energy 6
productivity in Pakistan Fan Zhang
47 Air Quality Impacts of Metro Rail in Mum- Working Paper Cropper and Suri 2022 Transport 5
bai
48 What are the Benefits of a Subway in Mum- Working Paper Cropper and Suri 2022 Transport 4
bai, India?
49 Balanced Growth and Public Capital: An Applied Economics Crowder and Himarios 1997 Cross-Sectoral 4
Empirical Analysis
50 Broadband Infrastructure and Economic The Economic Journal Czernich et al 2011 Digital 2
Growth
51 The Productivity of Public Capital: Journal of The Japanese and Daiji Kawaguchi, Fu- 2009 Cross-Sectoral 3
Evidence from the 1994 Electoral Reform International Economies mio Ohtake and Keiko
Tamada
iv
52 Road Improvement and Deforestation in the World Bank Policy Research Damania and Wheeler 2015 Transport 2
Congo Basin Countries Working Paper
53 Turnpike trusts and property income: New Economic History Review Dan Bogart 2009 Transport 1
evidence on the effects of transport
improvements and legislation in eighteenth
century England
54 Turnpike trusts and the transportation Explorations in Economic Dan Bogart 2005 Transport 3
revolution in 18th century England History
55 The Effects of electrification on employment IZA Journal of Labor and Dasso and Fernandez 2015 Energy 8
in rural Peru Development
56 Is Public Expenditure Productive? Journal of Monetary David Aschauer 1989 Cross-Sectoral 9
Economics
57 The effects of transportation networks on Columbia University David Canning and 1993 Transport 6
economic growth Marianne Fay
58 Infrastructure and education as instruments Economic Policy de la Fuentes and Vives 1995 Cross-Sectoral 1
of regional policy: evidence from Spain
59 The (Fuzzy) Digital Divide: The Effect of Working Paper De Stefano et al 2014 Digital 2
Broadband Internet Use on UK Firm
Performance
60 Broadband infrastructure, ICT use and firm Journal of Economic Behavior DeStefano et al 2018 Digital 3
performance: Evidence for UK firms & Organization
61 Long-term gains from electrification in rural The World Bank Economic Dominique Van de 2017 Energy 8
Review
India Walle, Martin Ravail-
lon, Vibhuti Mendiratta
and Gayatri Koolwal
62 Railroads of the Raj: Estimating the Impact American Economic Review Donaldson 2018 Transport 4
of Transportation Infrastructure
63 Road connectivity, population, and crop pro- Agricultural Economics Dorosh et al 2012 Transport 9
duction in Sub-Saharan Africa
64 Public-sector capital and the productivity The Review of Economics and Douglas Holtz-Eakin 1994 Cross-Sectoral 12
puzzle Statistics
65 How important are mobile broadband net- Information Economics and Edquist et al 2018 Digital 2
works for the global economic development? Policy
66 The regional returns of public investment World Development Eduardo Rodriguez- 2004 Cross-Sectoral 4
policies in Mexico Oreggia and Andres
Rodriguez-Pose
67 Infrastructure and Regional Economic New England Economic Eisner 1991 Cross-Sectoral; 5
Performance: Comment Review
Transport
68 Real government saving and the future Journal of Economic Behavior Eisner 1994 Cross-Sectoral 1
and Organization
v
69 Highway to Success: The Impact of the Economic Journal Ejaz Ghani, Arti Grover 2016 Transport 18
Golden Quadrilateral Project for the Location Goswami and William
and Performance of Indian Manufacturing Kerr
70 The effects of public capital on the Applied Economics Emanuela Marrocu and 2010 Cross-Sectoral 1
productivity of the Italian regions Raffaele Paci
71 Access to Markets and Rural Poverty: The Review of Economics and Emran and Hou 2013 Transport 1
Evidence from Household Consumption in Statistics
China
72 Productivity, Private and Public capital, and Applied Economics Letters Erenburg 1998 Cross-Sectoral 4
real wage in the US
73 Institutions, infrastructure, and economic Journal of Development Eco- Esfahani and Ramirez 2003 Energy; Digital 2
growth nomics
74 How much does infrastructure matter to Working Paper Estache et al. 2005 Transport; Energy; 3
Digital
growth in Sub-Saharan Africa?
75 Public capital and economic growth: a con- Journal of Economic Growth Etsuro Shioji 2001 Cross-Sectoral 12
vergence approach
76 Are government activities productive? The Review of Economics and Evans and Karras 1994 Cross-Sectoral; 5
Evidence from a panel of US states. Statistics Transport
77 Is Government Capital Productive? Evidence Journal of Macroeconomics Evans and Karras 1994 Cross-Sectoral 5
from a Panel of Seven Countries
78 Complementarities in Infrastructure: CEPR Discussion Paper Eynde and Wren-Lewis 2021 Energy; Transport; 3
Evidence from Rural India Cross-Sectoral
79 Road Development, Economic Growth, and IFPRI Research Report Fan and Chan-Kang 2005 Transport 1
Poverty Reduction in China
80 Public capital and economic performance: Giornale degli Economisti e Federico Bonaglia, 2000 Digital; Transport; 7
Evidence from Italy Annali di Economia Eliana la Ferrara and Cross-Sectoral
Massimiliano Marcelino
81 The impact of public capital and public in- Epge ensaios economicos Ferreira 1994 Cross-Sectoral 4
vestment on economic growth
82 Is all government capital productive? Federal Reserve Bank of Finn 1993 Transport 1
Richmond Economic Quarterly
83 Electricity Shortages and Firm Productivity: Journal of Development Eco- Fisher-Vanden et al 2015 Energy 1
Evidence from China’s Industrial Firms nomics
84 Powering up productivity: The Effects of NBER Working Paper Fiszbein et al 2022 Energy 4
Electrification on U.S. Manufacturing
85 Infrastructure and Private-Sector OECD Working Paper Ford and Poret 1991 Cross-Sectoral 1
Productivity
86 Productivity effects and determinants of The annals of regional science Fumitoshi Mizutani and 2010 Cross-Sectoral 4
public infrastructure investment. Tomoyasu Tanaka
vi
87 Should I Stay or Should I Go? The Rile of Journal of African Economies Gachassin 2013 Transport 2
Roads in Migration Decisions
88 Roads and Diversification of Activities in Development Policy Review Gachassin et al 2015 Transport 2
Rural Areas: A Cameroon Case Study
89 The contribution of publicly provided inputs Regional Science and Urban Garcia-Mila and 1992 Transport 1
to states’ economies Economics McGuire
90 The effects of public capital in state level Review of Economics and Garcia-Mila et al. 1996 Transport 2
production functions reconsidered Statistics
91 Airports, access and local economic Journal of Economic Gibbons and Wu 2020 Transport 6
performance evidence from China Geography
92 Urban growth and transportation Review of Economic Studies Gilles Duranton and 2012 Transport 4
Matthew Turner
93 Roads and trade: Evidence from the US Review of Economic Studies Gilles Duranton, Peter 2014 Transport 8
Morrow and Matthew
Turner
94 Subways and urban growth: Evidence from Journal of Urban Economics Gonzales-Navarro and 2018 Transport 4
earth Turner
95 Modern Telecommunications Infrastructure Industrial and Corporate Greenstein and Spiller 1995 Digital 2
and Economic Activity: An Empirical Change
Investigation
96 The Need for Speed: Impacts of Internet Working Paper Grimes et al 2009 Digital 1
Connectivity on Firm Productivity
97 Broadband access in the EU: An assessment Telecommunications Policy Gruber et al. 2014 Digital 1
of future economic benefits
98 Infraestructuras de transporte y Presupuesto y gasto público Gustavo Nombela 2005 Transport 5
productividad
99 Information communications technology Telecommunications Policy Haftu 2018 Digital 2
and economic growth in Sub-Saharan
Africa: A panel data approach.
100 Broadband adoption and firm productivity: Telecommunications Policy Haller and Lyons 2014 Digital 2
Evidence from Irish manufacturing firms
101 State infrastructure, the distribution of jobs, Working Paper Haughwout 2000 Transport 2
and productivity
102 La importancia de la infraestructura f´ısica en Estudios fronterizos Hećtor Barajas Bustillos 2012 Transport; Energy; 8
el crecimiento econo´mico de los municipios and Luis Gutierrez Flores Digital; Cross
de la frontera norte. Sectoral
103 Scale economies, returns to variety, and the Regional Science and Urban Holtz-Eakin and Lovely 1996 Cross-Sectoral 4
productivity of public infrastructure Economics
104 Spatial productivity spillovers from public International Tax and Public Holtz-Eakin and 1995 Cross-Sectoral 4
infrastructure: evidence from state high- Finance Schwartz
ways.
105 Infrastructure in a structural model of eco- Regional Science and Urban Holtz-Eakin and 1995 Transport 3
nomic growth Economics Schwartz
vii
106 Sectoral Value Added - Electricity World Bank Policy Research Hovhannisyan and 2021 Energy 6
Elasticities across Countries Working Paper Stamm
107 Infrastructure, Externalities, and Economic The World Bank Economic Re- Hulten et al 2006 Energy; Trans- 2
Development: A Study of the Indian view port
Manufacturing Industry
108 How Do Electricity Shortages Affect Indus- American Economic Review Hunt Allcott, Allan 2016 Energy 6
try? Evidence from India Collard-Wexler and
Stephen D. O’Connell
109 Heterogeneous effects of rural World Bank Policy Research Hussain Samad and Fan 2017 Energy 14
electrification: evidence from Bangladesh Paper Zhang
110 ICT adoption and wage inequality: evidence The World Bank Iacovone and Pereira- 2018 Digital 12
from Mexican firms Lo´pez
111 The employment and wage impact of broad- Canadian Journal of Eco- Ivus and Boland 2015 Digital 4
band deployment in Canada nomics
112 Infrastructure and growth in South Africa: World Development J.W. Fedderke and Z. 2009 Transport; Energy; 9
Direct and indirect productivity impacts of Bogetic´ Digital;
19 infrastructure measures Cross-Sectotal
113 Access to Markets and the Benefits of Rural The Economic Journal Jacoby 2000 Transport 4
Roads
114 Evaluating transport infrastructure projects Journal of Development Jason Russ, Claudia 2018 Transport 2
in low data environments: An application to Studies Berg, Richard Dama-
Nigeria nia, Alvaro Federico
Barra, Rubaba Ali and
John Nash
115 Impact of quantity and quality of Sustainability Javid 2019 Cross-Sectoral; 8
Infrastructure on economic growth in Energy
Pakistan: A disaggregated Analysis
116 Is public capital productive in Europe? International Review of Ap- Jerome Creel and Gwe- 2008 Cross-Sectoral; 3
plied Economics naelle Poilon Transport
117 Government capital and the production Economics Letters Jonathan Ratner 1983 Cross-Sectoral 2
function for US private output
118 Efectos de la infraestructura pu´blica sobre el Estudios de economia Jorge Rivera and 2004 Cross-Sectoral 1
crecimiento de la econom´ıa, evidencia para Patricia Toledo
Chile
119 Public capital, regional output, and Journal of Regional Science Jose Da Silva Costa, 1987 Cross-Sectoral 3
developments: some empirical evidence Richard Ellson and
Randolph Martin
120 Empirical Analysis of Transportation Transportation Joseph Berechman, Dil- 2006 Transport 3
Investment and Economic Development at ruba Ozmen and Kaan
State, County, and Municipality Levels Ozbay
121 Contribution of transportation investments to Transport Policy Kaan Ozbay, Dilruba 2007 Transport 3
county output. Ozmen-Ertekin and
Joseph Berechman
viii
122 Telecommunications infrastructure and eco- Applied Econometrics and Kala Seetharam Sridhar 2007 Digital 4
nomic growth: Evidence from developing International Development and Varadharajan
countries. Sridhar
123 New Estimates of Government Net Capital IMF Staff Papers Kamps 2006 Cross-Sectoral 26
Stocks for 22 OECD Countries, 1960 - 2001
124 Infrastructure Investment and Rural Eco- Growth and Change Kandilov and Renkow 2010 Digital 3
nomic Development: An Evaluation of
USDA’s Broadband Loan Program
125 Does Electrification Cause Industrial Working Paper Kassem 2021 Energy 18
Development? Grid Expansion and Firm
Turnover in Indonesia
126 Can digitization mitigate the economic dam- Telecommunications Policy Katz et al 2020 Digital 2
age of a pandemic? Evidence from SARS
127 The gains from market integration: The Mimeo - JMP Kebede 2021 Transport 4
welfare effects of new rural roads in Ethiopia
128 Infrastructure Productivity Estimation and Papers in Regional Science Kelejian and Robinson 1997 Cross-Sectoral; 4
its Underlying Econometric Specifications: Transport
A sensitivity Analysis
129 The contribution of local infrastructure to Logistics and Transportation Kemmerling and 2002 Cross-Sectoral 2
private productivity and its political Review Stephan
economy: evidence from the Canadian
goods-producing sector
130 Who Benefits Most from Rural World Bank Policy Research Khandker et al 2012 Energy 10
Electrification? Evidence in India Working Paper
131 Welfare Impacts of Rural Electrification: A Economic Development and Khandker et al 2013 Energy 18
Panel Data Analysis from Vietnam Cultural Change
132 Does infrastructure stimulate total factor The Quarterly Review of Eco- Khanna and Sharma 2020 Energy; Digital; 16
productivity? A dynamic heterogeneous nomics and Finance Transport
panel analysis for Indian manufacturing
industries
133 Benefits and Spillover Effects of Infrastruc- East Asian Economic Review Kijin Kim, Junkyu Lee, 2021 Energy; Digital; 12
ture: A Spatial Econometric Approach Manuel Leonard Albis Transport
and Ricardo III B. Ang
134 Flip the Switch: The Spatial Impact of the Journal of Economic History Kitchens and Fishback 2015 Energy 6
Rural Electrification Administration 1935 -
1940
135 Broadband and local growth Journal of Urban Economics Kolko 2012 Digital 4
136 The economic impact of broadband on Telecommunications Policy Koutroumpis 2009 Digital 2
growth: A simultaneous approach
137 Explaining High Transport Costs within World Bank Policy Research Lall et al 2009 Transport 1
Malawi: Bad Roads or Luck of Trucking Working Paper
Competition?
ix
138 Economic growth, telecommunications Telecommunications Policy Lam and Shiu 2010 Digital 2
development and productivity growth of
the telecommunications sector: Evidence
around the world
139 Highway Access and Human Capital ADBI Working Paper Li et al 2019 Transport 8
Investments in the Rural Regions of the
People’s Republic of China
140 Public Capital and Output Growth in Quarterly Review of Ligthart 2002 Cross-Sectoral 3
Portugal: An empirical Analysis Economics and Finance
141 Information and communication technology Pakistan Journal of Commerce Majeed and Ayub 2018 Digital 8
(ICT) and economic growth nexus: A and Social Sciences
comparative global analysis
142 Paving Streets for the Poor: Experimental Review of Economics and Marco Gonzales- 2016 Transport 3
Analysis of Infrastructure Effects Statistics Navarro and Climent
Quintana-Domeque
143 Spillovers and the Locational Effects of Journal of regional science Marlon Boarnet 1998 Transport 5
Public Infrastructure
144 Shaky roads and trembling exports: Journal of International Eco- Martincus and Blyde 2013 Transport 1
Assessing the trade effects of domestic nomics
infrastructure using a natural experiment.
145 Infrastructure and Productivity in the Regional Studies Mas et al. 1996 Cross-Sectoral 2
Spanish Regions
146 Competitividad, Productividad Industrial y Papeles de Economica Es- Mas Matilde, Joaquin 1993 Cross-Sectoral 2
Dotaciones de Capital Publico panola Maudos, Francisco
Pere´z, and Ezequiel
Uriel
147 Capital Publico y Productividad en las Re- Moneda y Credito Mas Matilde, Joaquin 1994 Cross-Sectoral 2
giones Espanolas Maudos, Francisco
Pere´z, and Ezequiel
Uriel
148 The impact of infrastructure expenditure Regional Studies Mehmet Akif Kara, 2016 Energy; Cross- 4
types on regional income in Turkey Seyhan Tas and Serkan Sectoral
Ada
149 The Effects of Roads on Trade and Migration NBER Working Paper Melanie Morten and 2018 Transport 4
: Evidence from a Planned Capital City Jacqueline Oliveira
150 Jobs! Electricity Shortages and World Bank Policy Research Mensah 2018 Energy 32
Unemployment in Africa Working Paper
151 Railroads and local economic development: NBER Working Paper Michael Haines and 2006 Transport 5
The United States in the 1850s Robert Margo
152 The effect of trade on the demand for skill - Review of Economics and Michaels 2008 Transport 3
Evidence from the Interstate Highway Statistics
System
x
153 Effect of public investment on the regional Review of Urban and Regional Mitsuhiko Kataoka 2005 Cross-Sectoral 4
economies in postwar Japan Development Studies
154 More Power to the People: Electricity The Journal of Economic His- Molinder et al 2021 Energy 1
Adoption Technological Change, and Labor tory
Conflict
155 Evidence on the complex link between International Journal of Devel- Moreno et al. 1997 Cross-Sectoral 3
infrastructure and regional growth opment Planning Literature
156 The Impact of public transport expansions Research in Transportation Moreno-Monroy and 2021 Transport 2
on informality: The case of the Sao Paulo Economics Ramos
Metropolitan Region
157 Why Has Productivity Growth Declined? New England Economic Re- Munnell 1990 Cross-Sectoral 2
Productivity and Public Investment view
158 Business Group Spillovers: Working Paper Naaraayanan and 2019 Transport 2
Evidence from the Golden Quadrilateral in Wolfenzon
India
159 Rural Roads, Poverty, and Resilience: World Bank Policy Research Nakamura et al 2019 Transport 2
Evidence from Ethiopia Working Paper
160 Did highways cause suburbanization? Quarterly Journal of Eco- Nathaniel Baum-Snow 2007 Transport 5
nomics
161 Roads, railroads and decentralization of The Review of Economics and Nathaniel Baum-Snow, 2017 Transport 7
Chinese cities Statistics Loren Brandt, J. Vernon
Henderson, Matthew A.
Turner, and Qinghua
Zhang
162 Does investment in national highways help Journal of Urban Economics Nathaniel Baum-Snow, 2020 Transport 8
or hurt hinterland city growth? Vernon Henderson,
Matthew Turner,
Qinghua Zhang and
Loren Brandt
163 ICT and economic growth–Comparing World Development Niebel 2018 Digital 2
developing, emerging and developed countries
164 Infrastructure, geographical disadvantage, World Bank Economic Review Nuno Limao and 2001 Cross-Sectoral 4
transport costs, and trade Anthony Venables
165 Public Capital and Private Sector Economic Record Otto and Voss 1994 Cross-Sectoral 2
Productivity
166 Is public capital provision efficient? Journal of Monetary Otto and Voss 1998 Cross-Sectoral 2
Economics
167 Public capital and private sector production Southern Economic Journal Otto and Voss 1996 Cross-Sectoral 2
in Australia
168 Public Investment and Economic Growth Applied Economics Otto and Voss 2003 Cross-Sectoral 1
169 An Empirical Study on public capital Applied Economics Owyong and 2001 Cross-Sectoral 2
spillovers from the USA to Canada Thangavelu
xi
170 Trade, Structural Transformation and Journal of Political Economy Pablo Fajgelbaum and 2022 Transport 7
Development: Evidence from Argentina Stephen Redding
1869 - 1914
171 Intertemporal output and employment The Economic Journal Panicos Demetriades 2000 Cross-Sectoral 13
effects of public infrastructure capital: and Theofanis Ma-
evidence from 12 OECD economies muneas
172 Infrastructure and economic growth in the World Bank Policy Research Paul Noumba Um, 2009 Transport; Digi- 3
Middle East and North Africa Paper Ste´phane Straub, and tal; Energy
Charles Vellutini
173 Transport Infrastructures, Spillover Effects Transport Reviews Pedro Cantos, Mer- 2005 Transport 10
and Regional Growth: Evidence of the Cedes Gumbau-Albert
Spanish Case and Joaquin Maudos
174 Productivity and Infrastructure in the Italian Giornale degli Economisti e Picci 1999 Cross-Sectoral 3
regions Annali di Economia
175 Determinants of long-run regional Regional Science and Urban Raffaello Bronzini and 2009 Transport; Cross- 8
productivity with geographical spillovers: Economics Paolo Piselli Sectoral
The role of R&D, human capital and public
infrastructure
176 The Average and Heterogeneous Effects of Journal of the European Eco- Remi Jedwab and 2022 Transport 6
Transportation Investments: Evidence from nomic Association Adam Storeygard
Sub-Saharan Africa 1960-2010
177 Growth of the Rails: Aggregate Productivity Mimeo Richard Hornbeck and 2021 Transport 9
Growth in Distorted Economies Martin Rotemberg
178 Assessing the contribution of public Revista de la CEPAL Roberto Urrunaga and 2012 Digital; Energy; 3
capital to private production: evidence from Carlos Aparicio Transport
the German Manufacturing sector
179 Evaluating China’s Road to Prosperity: A Regional Science and Urban Roberty et al 2012 Transport 6
new economic geography approach Economics
180 Telecommunications Infrastructure and Eco- American Economic Review Roller and Waverman 2001 Digital 2
nomic Development: A simultaneous
approach
181 Returns to local and transport infrastructure Regional Science Review Rosina Moreno and En- 2007 Transport 2
under regional spillovers rique Lopez-Bazo
182 Optimal Endowments of Public Investment: Economic Working Papers at Rubio, Roldan and 2002 Cross-Sectoral 1
An Empirical analysis for the Spanish Centro de Estudios Andaluces Garces
Regions
183 Electricity Provision and industrial Journal of Development Eco- Rud 2012 Energy 5
development: Evidence from India nomics
184 The Productivity Effects of Infrastructure: A Applied Economics Letters Rupika Khanna and 2021 Cross-Sectoral 12
Cross-Country Comparison Using Chandan Sharma
Manufacturing Industry Panels
185 Benefits of Electrification and the Role of World Bank Policy Research Samad and Zhang 2016 Energy 18
Reliability Paper
xii
186 The poverty impact of rural roads: Evidence Economic Development and Shahidur Khandker, 2009 Transport 12
from Bangladesh Cultural Change Zaid Bakht and Gayatri
Koolwal
187 Airports and urban sectoral employment Journal of Urban Economics Sheard 2014 Transport 8
188 Assessing the contribution of public International Review of Stephan 2003 Cross-Sectoral 2
capital to private production: evidence Applied Economics
from the German Manufacturing sector
189 Valuing rail access using transport Journal of Urban Economics Stephen Gibbons and 2005 Transport 3
innovations Stephen Machin
190 New road infrastructure: The effects on firms Journal of Urban Economics Stephen Gibbons, 2019 Transport 11
Teemu Lyytikaïnen,
Henry Overman and
Rosa Sanchis-Guarner
191 Is Public expenditure really productive: New Economic Modelling Sturm and de Haan 1995 Cross-Sectoral 2
Evidence for the USA and the Netherlands
192 Challenges to Mismeasurement Explana- Journal of Economic Syverson 2017 Digital 1
tions for the US Productivity Slowdown Perspectives
193 Broadband Internet and Household Welfare World Bank Policy Research Takaaki Masaki, 2020 Digital 12
in Senegal Paper Rogelio Granguilhome
Ochoa and Carlos
Rodriguez-Castelan
194 Public capital and private sector Federal Reserve Bank of St Tatom 1991 Cross-Sectoral 1
performance Louis Review
195 Economic Impacts of Mobile Versus Fixed Telecommunications Policy Thompson and Garbacz 2011 Digital 2
Broadband
196 Impact of information and Sustainability Toader et al 2018 Digital 8
communication technology infrastructure on
economic growth: An empirical assessment
for the EU countries
197 Public Capital and Private Productivity Review of Economics and Vijverberg et al. 1997 Cross-Sectoral 2
Statistics
198 The impact of telecoms on economic growth Working Paper Waverman et al. 2005 Digital 1
in developing countries
199 Heterogeneous Effects of Inter- and Intra- Cambridge Journal of Regions, Yang Chen, Nimesh Sa- 2018 Transport 4
city Transportation Infrastructure on Eco- Economy and Society like, Fushu Luan and
nomic Growth: Evidence from Chinese Ming He
Cities
200 Public infrastructure investment, economic Proceedings of the 2012 Inter- Yu Nannan and Mi 2012 Cross-Sectoral 1
growth and policy choice: Evidence from national Conference on Public Jianing
China. Management (ICPM 2012)
201 Does infrastructure have a transitory or Economic Modelling Zhang and Ji 2017 Digital; Energy; 4
longer-term impact? Evidence from China Transport
xiii
B Variables in the Dataset
We have compiled a rich dataset containing the following information:
• Identifiers: Title of the Paper, Author(s), Source, Year, Type of Publication (Peer-reviewed
or other), Author(s) affiliation (Academia, Government, International Institution, Private)
• Sample and Data: Names of countries covered in the study; Number of countries covered
in the study, Indicator for cross-country study, Indicator for developing country only sample,
Indicator for developed country only sample, Region, Income Group of the countries
covered in the study, Time period covered, Type of data used (Country-level, Region-level,
district/municipality-level, household-level, firm-level), Indicator for the use of Spatial data
in the analysis
• Framework: Indicator for the framework used in the analysis (General Equilibrium Trade
model; Production Function; Cost Function; Other Structural Model)
• Empirical Strategy: Indicator for whether the paper treats endogeneity; Technique used to
address potential endogeneity (IV, GMM, RCT, RD, DID, Fixed-effects), Indicator for
cointegration, Indicator for using long-difference models
• Outcome and Treatment: Dependent variable, Broad category of the dependent variable
(Output-Micro, Output-Macro, Labor), Independent variable (treatment), Broad category of
the independent variable (Digital, Energy, Transport, Cross-Sectoral), Coefficient reported
in the paper, Coefficient used in the analysis (semi-elasticities converted to elasticities and
sign of treatment effect corrected depending on the definition of the independent variable),
Standard Error, Level of Statistical Significance, Number of Observations, t-statistic, Link
between Infrastructure and Development (Positive or Negative)
xiv

WBWP Infra

Uploaded by

Document Information

Original Description:

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

WBWP Infra

Uploaded by

Copyright:

Available Formats

Public Disclosure Authorized

Policy Research Working Paper 10350

The Impact of Infrastructure

Infrastructure Chief Economist Office

Produced by the Research Support Team

JEL: H54, O18, O47, Q40, R40

Keywords: Infrastructure, Transport, Energy, ICT, Meta-analysis.

This study contributes to the literature on infrastructure by providing a systematic assessment

2.1 Studies included in the review: Search and selection criteria

• Transport, comprising roads, railroads, ports, and airports.

• Energy, covering production, transmission, and distribution of electricity to households and

Yit = α + δXit + θi + θt + εit (1)

• Output, which includes indicators of production, income, expenditures, etc., in levels or

• Labor market, including indicators of employment and wages.

• Inequality and poverty, such as Gini or poverty share indicators.

• Population, including levels as well as migration flows.

2.2 Descriptive statistics

Econometrically, the corresponding test relies on the following equation:

̂ 𝑖𝑠 = β0 + β1S Eis + εis

̂ 𝑖𝑠 = β0 + β1S Eis2 + εis

Equation 2 thus becomes:

̂ 𝑖𝑠 = β0 + Σ βk Mkis + β1S Eis + υis + ϵis

4.1 Funnel plots

4.2 FAT-PET-PEESE estimations

4.3 Robustness to outliers

4.4 Specific sub-sectors

5 Discussion and conclusion

elasticity, Y is GDP, and Inf is the stock of infrastructure.

Aschauer, D. A. (1989). Is public expenditure productive? Journal of monetary economics,

Greenstein, S. (2020). The basic economics of internet infrastructure. Journal of Economic

Redding, S. J. and Rossi-Hansberg, E. (2017). Quantitative spatial economics. Annual Review of

Stanley, T. D. and Doucouliagos, H. (2012). Meta-regression analysis in economics and business.

Funnel plots by type of infrastructure

Leave-one-out robustness plots

Source: Authors’ elaboration.

Figure 2: Funnel plot - Cross-sectoral studies

Source: Authors’ elaboration.

Source: Authors’ elaboration.

Figure 4: Funnel plot - Energy

Source: Authors’ elaboration.

Source: Authors’ elaboration.

Figure 6: Funnel plot - Output macro

Source: Authors’ elaboration.

Source: Authors’ elaboration.

Figure 8: Funnel plot - Labor

Source: Authors’ elaboration.

Figure 10: Funnel plot - Roads

Source: Authors’ elaboration.

Source: Authors’ elaboration.

((a)) Sector-level samples

((b)) Sector-outcome-level Samples

Tables 1 to 3 present descriptive statistics.

Cross-sectoral Digital Energy Transport Total

Table 2: Descriptive statistics: Sign of estimates by sectors

Cross-sectoral Digital Energy Transport Total

(1) (2) (3) (4) (5)

1/se 0.0166 0.1624*** 0.0138 -0.0004 0.0230

Observations 991 232 145 219 395

Table 5: PEESE estimates

(1) (2) (3) (4) (5)

1/se 0.0200 0.1613*** 0.0152 0.0065** 0.0311**

Observations 991 232 145 219 395

(1) (2) (3) (4) (5) (6) (7) (8)

1/se 0.0200 0.1613* 0.0152 0.0065 0.0311**

1/se 0.0337* 0.0132 -0.0012 0.0069 0.0000 -0.0053 0.0433* 0.0202

1/se 0.0429* 0.0151 0.0062 0.0412* 0.0021* 0.0104 0.0474* 0.0446

se 0.1274 59.9045 9.2492* 20.5807* 7.8994* 4.4016** 7.9751* 13.4962**

1/se 0.0557* 0.0048 0.0057 0.0442* 0.0015* 0.0104 0.0203* 0.0463

se 0.1008 66.3753* 9.2573* 20.5037* 7.9128* 4.4016 8.8426 13.3924**