ArticlePDF Available

Abstract and Figures

Universities in developing countries have rarely been able to subscribe to academic journals in the past. The “Online Access to Research in the Environment” initiative (OARE) provides institutions in developing countries with free online access to more than 11,500 environmental science journals. We analyze the effect of OARE on (1) scientific output and (2) scientific input as a measure of accessibility in five developing countries. We apply difference-in-difference-in-differences estimation using a balanced panel with 249,000 observations derived from 36,202 journal articles published by authors affiliated with 2,490 research institutions. Our approach allows us to explore effects across scientific fields, i.e. OARE vs. non-OARE fields, within institutions and before and after OARE registration. Variation in online access to scientific literature is exogenous at the level of scientific fields. We provide evidence for a positive marginal effect of online access via OARE on publication output by 29.6% with confidence interval (18.5%, 40.6%) using the most conservative specification. This adds up to 2.07 additional articles due to the OARE program for an average institution publishing 7.0 articles over the observation period. Moreover, we find that OARE membership eases the access to scientific content for researchers in developing countries, leading to an increase in the number of references by 8.4% with confidence interval (5.6%, 11.2%) and the number of OARE references by 14.5% with confidence interval (7.5%, 21.5%). Our results suggest that productive institutions benefit more from OARE and that the least productive institutions barely benefit from registration.
Content may be subject to copyright.
Contents lists available at ScienceDirect
Research Policy
journal homepage: www.elsevier.com/locate/respol
Does online access promote research in developing countries? Empirical
evidence from article-level data
Frank Mueller-Langer
a,b,c,
, Marc Scheufen
d,e
, Patrick Waelbroeck
f,g,#
a
European Commission, Joint Research Centre, Digital Economy Unit, Seville, Spain
b
Max Planck Institute for Innovation and Competition, Munich, Germany
c
Bundeswehr University Munich, Department of Business Administration, Munich, Germany
d
Ruhr University Bochum, Bochum, Germany
e
German Economic Institute (IW), Cologne, Germany
f
Télécom ParisTech, Paris, France
g
CESIfo, Munich, Germany
ARTICLE INFO
Keywords:
Online access
Scientific productivity
Access to knowledge as publication input
Difference-in-difference-in-differences (DDD)
estimation
Instrumental variables (IV) estimation
Bayesian Markov-Chain-Monte-Carlo (MCMC)
estimation
JEL codes:
L17
O33
ABSTRACT
Universities in developing countries have rarely been able to subscribe to academic journals in the past. The
“Online Access to Research in the Environment” initiative (OARE) provides institutions in developing countries
with free online access to more than 11,500 environmental science journals. We analyze the effect of OARE on
(1) scientific output and (2) scientific input as a measure of accessibility in five developing countries. We apply
difference-in-difference-in-differences estimation using a balanced panel with 249,000 observations derived
from 36,202 journal articles published by authors affiliated with 2,490 research institutions. Our approach
allows us to explore effects across scientific fields, i.e. OARE vs. non-OARE fields, within institutions and before
and after OARE registration. Variation in online access to scientific literature is exogenous at the level of sci-
entific fields. We provide evidence for a positive marginal effect of online access via OARE on publication output
by 29.6% with confidence interval (18.5%, 40.6%) using the most conservative specification. This adds up to
2.07 additional articles due to the OARE program for an average institution publishing 7.0 articles over the
observation period. Moreover, we find that OARE membership eases the access to scientific content for re-
searchers in developing countries, leading to an increase in the number of references by 8.4% with confidence
interval (5.6%, 11.2%) and the number of OARE references by 14.5% with confidence interval (7.5%, 21.5%).
Our results suggest that productive institutions benefit more from OARE and that the least productive institu-
tions barely benefit from registration.
1. Introduction
While global online access has laid the groundwork for involving all
nation-states in science, universities in developing countries have rarely
been able to subscribe to academic journals in the past (Annan, 2004).
For instance, most libraries in Sub-Saharan African countries had no
access to any scientific journal for years (Suber and
Arunachalam, 2005). The Online Access to Research in the
Environment (OARE) initiative seeks to provide free or reduced-fee
online access for researchers of registered institutions in the field of
environmental science. It was launched by the United Nations En-
vironment Programme (UNEP) and Yale University in October 2006.
We focus our analysis on the five developing countries that are most
active in terms of both publishing (number of articles in Web of Sci-
ence) and registration with the OARE initiative: Kenya and Nigeria
(Sub-Saharan Africa) and Bolivia, Ecuador and Peru (South America).
https://doi.org/10.1016/j.respol.2019.103886
Received 9 January 2019; Received in revised form 9 October 2019; Accepted 4 November 2019
Corresponding author at: European Commission, Joint Research Centre, Digital Economy Unit, Edificio Expo, Calle Inca Garcilaso 3, 41092 Seville, Spain
E-mail address: frank.mueller-langer@ip.mpg.de (F. Mueller-Langer).
#
We are grateful to Kimberly Parker and the World Health Organization for providing us with data on registration dates of the institutions that joined the
Research4Life initiatives. We thank Luis Aguiar, Néstor Duch-Brown, Estrella Gomez-Herrera, Dietmar Harhoff, Stan Liebowitz, Bertin Martens, Fabio Montobbio,
Laura Rosendahl Huber, Joel Waldfogel, Richard Watt, Michael Weber and the participants of the 2016 MPI-IC research seminar, the 2016 Workshop on “The
Organization, Economics and Policy of Scientific Research”, the 2015 MPI-IC guest lecture series and the 2015 SERCI Conference for comments. Research assistance
was provided by Sebastian Osterrieth, Laura Sundsgaard and Christoph Winter. The views expressed are purely those of the authors and may not in any circumstances
be regarded as stating an official position of the European Commission.
Research Policy 49 (2020) 103886
0048-7333/ © 2019 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license
(http://creativecommons.org/licenses/BY/4.0/).
T
We investigate the impact of the OARE initiative on both local pub-
lication output as well as publication input. As for the input analysis, we
explore the effect of online access to OARE journals on the number of
references in general and the number of references to articles published
in OARE journals.
All developing countries are eligible, but the initiative distinguishes
between so-called Group A and Group B countries. Registered research
institutions in Group A countries (gross national income (GNI) per ca-
pita below $1,600) receive free online access to all journals that are
available under the OARE initiative whereas institutions in Group B
countries (GNI per capita below $5,000) receive access for a reduced
fee of $1,000 per year. Notably, the OARE registration date varies by
institution, i.e., we have different pre-and post-OARE cutoffs for dif-
ferent OARE member institutions.
Using bibliometric article-level data from Web of Science (WoS) and
OARE registration data from January 2000 to June 2012, we analyze
the impact of OARE on the publication output (i.e. number of articles)
and input (i.e. number of references) of research institutions. Our
identification strategy is based on the fact that OARE limits free or
reduced-fee online access to environmental (OARE) journals. We use a
difference-in-difference-in-differences (DDD) estimation that explores
differences in publication output and input across OARE and non-OARE
fields within institutions that registered with the OARE initiative and
those that did not – before and after joining OARE. The underlying idea
is that only researchers working on environmental issues can be im-
pacted by free or reduced-fee online access to OARE journals after an
institution has registered with OARE. In contrast, other scientific fields
within the same institution (and OARE fields within the same institu-
tion but before OARE registration) will not benefit from the OARE
program. Using this methodology, we compare the publication output
(i.e., published journal articles) and publication input (i.e., references
cited in these articles) in OARE fields in a given OARE member in-
stitution with the output and input in other fields at the same institution
and with the output and input of non-member institutions – before and
after joining the initiative.
The DDD method has the advantage that it can deal with concerns
regarding self-selection at the institutional level that could imply that
more productive universities might be more likely to register with
OARE. We mitigate these concerns by looking at OARE and non-OARE
disciplines in the same institution. In addition, we use instrumental
variables and Bayesian methods to account for potential unobserved
endogeneity. We find that OARE membership increases the overall
number of publications of a research institution by +29.6% (95%-CI:
18.5% to 40.6%), i.e. OARE membership adds, in average, 2.07 (95%-
CI: 1.3 to 2.8) articles per institution over the whole period of 50
quarters. This is equivalent to 0.04 extra articles per quarter for an
average institution, a rather small economic effect. However, the most
productive institutions benefit most from OARE membership while the
least productive institutions barely benefit from registration. These
findings may have important policy implications as a higher pro-
ductivity level in academic research may have a positive effect on
economic growth and other means of economic prosperity, e.g. en-
vironmental innovation (Romer, 1986 &1990;Griliches, 1957 &1992;
Jaffe, 1989).
1
Moreover, our results suggest that OARE membership
promotes access to academic knowledge as an input in the scientific
production, measured by the number of references cited by researchers
in our dataset. We find that researchers in OARE fields within OARE-
registered institutions cite 8.4% (95%-CI: 5.6% to 11.2%) more journal
articles than other researchers and also cite 14.5% (95%-CI: 7.5% to
21.5%) more articles from OARE journals than other researchers, re-
spectively.
The remainder of the paper is organized as follows: Section 2 relates
our work to the literature on the economics of science, innovation and
economic growth. In Section 3, we describe the OARE initiative, discuss
potential issues related to self-selection and endogeneity and provide an
overview of the data and the variables under study. Section 4 describes
the methodology. In Section 5, we present the results of our empirical
analysis, provide extensions and discuss robustness checks. Section 6
concludes.
2. Related literature
Access to scientific research has recently attracted widespread in-
terest from economics scholars (Furman and Stern, 2011;McCabe and
Snyder, 2015;Sorensen, 2004) and policy-makers (European
Commission, 2012). In particular, open access (OA) has been subjected
to a broad discussion on whether it is a promising new business model
in the digital economy (Suber, 2012;Scheufen, 2015;McCabe and
Snyder, 2018).
2
The literature on open access can broadly be structured along three
lines of research: studies investigating the effects of different publishing
models (Shavell, 2010
3
;Jean and Rochet, 2010); studies analyzing the
impact of different publishing models on readership and citations
(Gaulé and Maystre, 2011;McCabe and Snyder, 2014,2015;Mueller-
Langer and Watt, 2018); and studies directed towards a scientist's at-
titude and behavior regarding OA publishing (Hanauske et al., 2007;
Eger et al., 2015). Our paper seeks to contribute to the first line of
research. In particular, we study the effects of a change in the ability of
researchers in developing countries to access academic works. We
analyze the impact of this change before and after these researchers’
institutions joined the OARE initiative, and we compare the results to
those disciplines within institutions for which the access mode re-
mained unchanged over time. Our research discusses the impact of free
or reduced-fee online access on scientific production in developing
countries, for which we find little prior literature.
4
However, the need
for such research is emphasized by Annan (2004). Our DDD approach
allows us to examine the effect of OARE controlling for article char-
acteristics and institutional characteristics such as rank, city population
and the distance to the largest domestic city. Evans and Reimer (2009a)
emphasize the need to further assess the role of open access in devel-
oping countries. Evans and Reimer (2009b, p. 5) show that “lower-
middle-income countries tend to much more frequently cite freely
available journals, but the poorest countries do not.” Thus, scientists in
the poorest countries seem to have virtually no access to online content.
1
We will elaborate on the link between academic research and economic
growth in Section 2.
2
Two arguments mainly drive this debate. First, with the advent of the
Internet and the development of technologies to digitize information goods,
scientific journal publishers have found new means to price discriminate (“big
deals”), which has led to a sharp increase in journal subscription prices
(Bergstrom and Bergstrom, 2004;Ramello, 2010) and hence higher costs of
access to research. In contrast, OA provides free and unrestricted access to
academic works (McCabe and Snyder, 2005,2014). Second, the copyright
system that is behind these pricing schemes is built on the idea that commercial
exclusivity granted by copyright generates the main incentive for the creator of
a copyright work. Researchers, in contrast, are primarily motivated by re-
putation rather than by financial gains. Especially for journal articles, authors
typically do not receive any royalties, since the copyright is generally trans-
ferred to the publisher. Some authors even argue that an abolishment of
copyright and hence a forced OA regime would foster scholarly esteem
(Shavell, 2010).
3
Shavell (2010) argues that (a) readership is higher under open access, (b) a
higher readership increases scholarly esteem, (c) research institutions would
bear the costs of a shift towards the “author pays” model and (d) there are
several reasons why legal action is necessary to facilitate a change towards a
universal OA regime. Several researchers have critically assessed the assump-
tions made in Shavell (2010). See Mueller-Langer and Scheufen (2013) for a
review.
4
Gaulé (2009),Frandsen (2009) and Davis (2011) are notable exceptions that
we will discuss in this section.
F. Mueller-Langer, et al. Research Policy 49 (2020) 103886
2
Evans and Reimer (2009a) suggest that poor infrastructure and slow
internet access may explain this difference in citation rates.
McCabe and Snyder (2015) criticize their paper, arguing that Evans and
Reimer (2009a) do not control for citation trends.
5
Our approach
complements the two papers, as we analyze both input and output
trends of access to academic works for researchers in the developing
world.
6
We contribute to this strand of literature by investigating the
role of free and reduced-fee online access in developing countries on
scientific output and input. Our paper is also related to Gaulé (2009),
Frandsen (2009) and Davis (2011).Gaulé (2009) and Frandsen (2009)
explore the access restrictions to scientific literature of developing
country researchers as compared to developed country researchers.
Using a database of 43,150 scientific papers published by Swiss and
Indian researchers in 2007, Gaulé (2009) compares backward citation
patterns for Swiss and Indian scientists to analyze the difficulties faced
by developing country researchers in accessing scientific literature.
Gaulé (2009) finds that Indian authors have shorter reference lists and
are more likely to cite articles that appeared in open access journals. In
a similar vein, Frandsen (2009) conducts a publication and citation
analysis for the use of OA journals (for 451 journals) in biology by
researchers in developing countries. Frandsen's (2009) results suggest
that researchers from developing countries are not attracted to open
access journals to a larger extent than researchers from developed
countries.
Our paper is closely related to Davis (2011) that analyzes the
impact of a digital collection of journal articles in agriculture and
allied subjects (TEEAL) on (1) the number of produced articles, (2)
the number of references (in general) and (3) the number of TEEAL
references for researchers in eleven developing countries. While
Davis (2011) finds no effect on the production output of subscribed
institutions, he provides evidence for a significant impact on the
number of references both in general and for the TEEAL journals.
However, our approach differs from Davis (2011)’s approach in
several aspects. First, we investigate the impact of free and reduced-
fee online access to scientific works for member versus non-member
institutions, while Davis (2011) explores the effect of offline access
to scientific works. In this regard, it is particularly noteworthy that
the TEEAL journal content is shipped to subscribing institutions on a
portable hard-drive copy (CD-ROM) upon payment of $ 5,000.
7
As a
result, both the offline access and the payment may be seen as
substantial costs in making use of TEEAL journals for researchers in
the developing world. Secondly, the OARE initiative offers access to
more than 11,500 journals in environmental sciences, while TEEAL
offers access to over 200 journals in agricultural sciences. Thirdly,
we broaden the geographical scope by looking at Sub-Saharan
African and South American countries. Finally, we do not only look
at the differences in impact for member versus non-member in-
stitutions, but also explore the effect within institutions by applying
a triple difference approach to investigate both input and output
effects. Similar to Davis (2011), we also find that free or reduced-fee
access to academic journals increases the number of references both
in general as well as with respect to the initiative (TEEAL in
Davis (2011), OARE in our analysis).
Our paper also contributes to the literature in the broader field of
economics of science and innovation investigating the role of science
and scientific research in the advancement of technologies and hence in
fostering economic growth (Dasgupta and David, 1994;Dosi, 1988;
Merton, 1973;Murray et al., 2016).
8
In general, Romer (1986,1990)
highlights the role of academic research as a major factor for techno-
logical innovations and hence for economic growth. Before Romer, the
literature especially by Solow (1956) and Swan (1956) were able to
explain the role of academic research for economic growth by means of
a residual as growth was exogenously determined. Romer's endogenous
growth theory emphasizes the relevance of spill-overs from academia.
Accordingly, when free online access increases scientific output, this
eventually may have a positive effect on innovation and economic
growth. Extending on Romer (1990) several authors have emphasized
the importance of knowledge spillovers from science for economic
growth (Griliches, 1992;Jaffe, 1989;Audretsch and Feldman, 1996;
Acs et al., 1994).
9
However, these spillovers cannot be taken for
granted as we find that only 5 percent of eligible institutions are OARE
members which points to the unused potential of the initiative.
3. Data and variables
3.1. The OARE initiative
The OARE initiative is led by the United Nations Environment
Programme (UNEP) in partnership with major publishers in environ-
mental science.
10
OARE was launched in October 2006. Today, OARE
offers access to more than 11,500 peer-reviewed scientific journal titles
published by 461 OARE partners in more than 100 eligible countries. In
this regard, eligibility distinguishes between Group A (free online ac-
cess) and Group B (low-cost access) countries,
11
depending on the
countries’ GNI per capita. Institutions in countries with a GNI per capita
at or below $ 1,600 receive free access to the full content of journal
articles, while institutions in countries with a GNI per capita below $
5,000 pay a fee of $ 1,000 per year. However, institutions have to
register to OARE in order to receive access. In this respect, OARE offers
courses and workshops for librarians and researchers in order to make
the initiative known to a wider audience.
Fig. 1 illustrates the rate of adoption of OARE over time (quarters)
in all countries (as given by the middle solid line) and separately in
Group A countries (upper dashed line) and Group B countries (lower
dashed line).
12
The rate of adoption is measured by the cumulative
number of institutions that joined OARE in a given quarter divided by
the total number of institutions, i.e. 2,490 institutions in all countries
(Group A countries: 1,599; Group B: 891). Finally, it is worth noting
that about 5% of all eligible research institutions in Group A countries
and about 4% of all eligible research institutions in Group B countries
had joined OARE in the last quarter under study (June 2012).
13
To account for self-selection, we apply a triple difference-in-differ-
ence approach dealing with concerns regarding self-selection at the
institutional level that could imply that more productive universities
5
We follow McCabe and Snyder's (2015) approach to control for trends in
publication output.
6
Input is measured by the relative number of cited OARE articles in a given
article, while output is measured by the total number of articles of a given
institution.
7
We obtained the information on the fee payment from the TEEAL website at
https://teeal.org/purchase.
8
See also Stephan (1996) for an overview of the economics of science lit-
erature.
9
See Diamond (1994) for an overview on Zvi Griliche's contributions for
understanding the economics of technology and growth. See also
Geroski (2000),Hall (2004),Hall and Kahn (2003) and Mansfield (1961,1963).
10
See http://oare.research4life.org/content/en/partners.php for an overview
on the major partners of the OARE initiative.
11
Please note that countries can also convert from one group to the other if
the GNI per capita changes over time. As such, Bolivia changed from group A to
group B in 2017. For the time horizon under study, however, we do not find any
group transitions.
12
We used Internet Archive's Wayback Machine to explore possible group
changes over time. All countries under study remained in the same group for
the period under study, i.e. 2000 to 2012.
13
Note that the total number of eligible institutions refers to institutions that
have observable research output in the period under study. We exclude non-
research institutions from our analysis, i.e., we drop institutions that did not
publish any journal article during the period under study.
F. Mueller-Langer, et al. Research Policy 49 (2020) 103886
3
might be more likely to register with OARE. The triple difference-in-
difference approach compares the treatment effect in OARE field arti-
cles with a control group of non-OARE field articles at the institution
level, thus mitigating self-selection of institutions into the OARE in-
itiative.
3.2. Instrumental variables and endogeneity
There might still be endogeneity issues if the unobserved variable is
correlated with the OARE treatment effect. There are two problems that
we may worry about: (1) unobserved endogenous benefits (not con-
trolled for by other independent fixed effects, such as institutions
having better digital infrastructures) which could result in self-selection
into OARE; (2) unobserved endogenous information problems resulting
from the fact that only well-informed institutions can join the OARE
initiative.
We extend our analysis by using an instrumental variable (IV) ap-
proach that takes into account the fact that many institutions located in
similar areas in a given country joined other online-access initiatives in
addition to OARE (see also Section 4.2. “DDD using Instrumental
Variables” and Section 5.2. “The Effect of OARE using IV Estimation”).
In particular, next to the OARE initiative there are further initiatives
within the “Research4Life”-program initiated to foster access to scien-
tific content also in health science (HINARI) and agricultural science
(AGORA). The Health Inter-Network Access to Research Initiative
(HINARI) was launched in January 2002 (quarter 9), offering free or
reduced-fee access to more than 15,000 journals in biomedical and
health research. The Access to Global Online Research in Agriculture
(AGORA) initiative was launched in October 2003 (quarter 16), pro-
viding with access to a collection of around 14,500 key journals in
agricultural science for researchers in developing countries.
Using information on these Research4Life initiatives, we implement
our IV approach using two methodologies. First, we use a two-stage
least squares method that initially analyses the factors that explain self-
selection into the OARE initiative using instruments based on the
average number of institutions that joined OARE, HINARI and AGORA
in similar areas and then uses the predicted value of the treatment effect
in the productivity equation. Secondly, we simultaneously estimate the
parameters of the selection and the productivity equation using
Bayesian MCMC methods (see also the methodology Section 4.3. and
the results Section 5.3.).
3.3. Data
Our dataset is built from three main sources. First, we collected
bibliometric article-level data from WoS for the five countries under
study. We focus our analysis on five countries for the following reasons.
On the one hand, we choose the most productive countries in terms of
the total number of research articles from January 2000 to June 2012
for both geographical regions (Sub-Saharan Africa and South America).
On the other hand, we look at countries that exceed a threshold of at
least 20 OARE institutions in order to have variation across institutions
within countries.
14
Second, we gathered institutional data including
institutions’ registration with OARE. Third, we extracted the rank of the
institutions from the Ranking Web of Universities.
Regarding the first data source, we collected a panel dataset con-
taining metadata for 36,202 research articles. The period under study
starts in January 2000 (quarter 1) and ends in June 2012 (quarter 50).
We obtain article metadata from WoS. The WoS data contain the in-
stitutions of the authors, the title of the paper, journal information
(publication date, number of pages, volume number, issue number) and
the number of citations. Overall, we have 2,490 institutions that pub-
lished at least one article over the period under study.
We use article-level data for assigning different characteristics to
each single article, accounting for the field of research, institutional
affiliations of the authors, cooperation with authors from outside the
developing world and other controls such as number of references,
pages etc. Since the OARE initiative offers free or reduced-fee online
access to research in environmental science, we create a dummy vari-
able indicating whether an article falls under an OARE research area.
We define an article as falling under an OARE research area if its
“Research Area” provided by WoS also appears frequently in the titles
of OARE journals. We proceed as follows. First, for all articles under
study, we extract all terms from the WoS “Research Area” field. Second,
we order these research area terms by frequency, i.e., we count how
many articles in the data fall under a given single-word term (hence-
forth, WoS research area terms). For instance, in the case of articles of
authors affiliated with Nigerian universities, the term “environmental”
Fig. 1. OARE ADOPTION OVERALL AND BY COUNTRY
GROUP Adoption patterns of OARE for all countries (as
given by the middle solid line) and by country group (da-
shed lines). The OARE initiative started in quarter 28.
Quarter 28 is thus the earliest possible quarter for OARE
registration. Group A countries: Kenya, Nigeria and
Bolivia. Group B countries: Ecuador and Peru. Fig. 1 sug-
gests that different institutions registered with OARE at
different points in time. This implies that we have multiple
cut-offs for before and after OARE depending on the re-
spective institution. Fig. 1 also implies that only about 5%
of all eligible institutions joined OARE after a period of 5 ½
years. It points to the unused potential of the OARE in-
itiative.
14
Please also note that the data creation process involved manual matching
of institutions using different versions of search terms in Stata's string matching
functions. We will further elaborate on the data creation process below in
Section 3.3.
F. Mueller-Langer, et al. Research Policy 49 (2020) 103886
4
appears 2,179 times, whereas the term “architecture” appears once.
Next, we extract the 200 most frequent terms that appear in the com-
plete list of titles of OARE journals (henceforth, top 200 title terms).
Matching these two lists (WoS research area terms and top 200 title
terms), we obtain the top 50 OARE research areas. The top 50 OARE
research areas are given by the 50 most frequent WoS research area
terms that are also included in the top 200 title terms. Finally, we use a
“top 50 OARE research”-dummy (which is one if an article falls under
the top 50 OARE research fields, 0 otherwise). Distinguishing between
OARE fields and non-OARE fields within a given institution allows us to
explore the effect of online access to OARE journals on scientific output
in OARE fields before and after OARE registration as well as in OARE
fields as compared to non-OARE fields before and after OARE regis-
tration.
15
Our sample contains all articles of researchers of the countries under
study, including both single and multiple authored articles. However,
dividing the share of a publication between different (local) authors to
determine the respective contributions of authors is a challenging task
for at least two reasons. First, there is no consensus within and across
disciplines on how to account for multiple authorships. In particular,
taking each author of a paper fully into account would overestimate the
output produced. Creating a weight for multiple authored papers by
dividing each publication by the number of authors, however, would
also necessarily involve assumptions on the habits of co-authorship. In
some disciplines (or publishing cultures), the order of authors has clear
implications. Sometimes the first author or the last author is perceived
as the “main author” of a research article. Other disciplines choose the
order of authors alphabetically or by status. All of this makes it hard to
operationalize multiple-authored papers from one country. Second, to
the best of our knowledge, McCabe and Snyder (2015) is the only re-
ference that discusses the issue of single versus multiple authors with
respect to online access. They nevertheless restrict their sample to
single authors (from a local country) due to the difficulties in dividing
the share of a multiple authored paper between the authors. However,
to consider multiple co-authored articles in addition to single authored
articles has two main advantages. First, only looking at single authored
articles would substantially reduce our sample by 18,955 articles, that
is, more than 50% of our dataset. Second, multiple co-authored articles
may have different characteristics than single authored articles.
In the light of these advantages, we address these concerns as fol-
lows. First, we account for multiple authorship by simply dividing the
institutional share of each paper by the number of authors. For in-
stance, a paper with two authors from two institutions leads to an in-
crease in output of 0.5 for each of these institutions.
16
Second, to test
the robustness of our results, we provide the regression results for single
authored papers in Section 5.5.2 (Table 9).
17
The results are remarkably
similar.
To construct the (balanced) panel, we gather article level informa-
tion by institution, field (OARE vs. non-OARE) and quarter for each
country under study. For each country, we then merge rank and city
information – including population and distance data – from separate
datasets. Subsequently, we merge all individual country data into one
dataset.
18
We distinguish country-specific information by generating a
unique country ID for all countries. In total, we obtain 249,000
institution-discipline-quarter triplets, which constitute our unit of ob-
servation.
In assigning institutions to authors of articles from the countries
under study, we use Stata string-matching functions, searching for
snippets of institution names and abbreviations. In particular, we
manually account for different versions and spellings of institutions, as
WoS does not provide with a unique number or code to unambiguously
identify a particular institution of interest. Most importantly, also
spelling errors, case sensitivity as well as abbreviations impede an au-
tomatic matching of articles and institutions. Last but not least, we
repeat the string matching process for each country file for each author
level, accounting for up to 11 levels (11 authors for each article) and
distinguishing different author level IDs. We unambiguously identify
459 research institutions that are part of the Ranking Web of World
Universities and/or OARE member institutions, forming the core uni-
versities for the string-matching process.
19
For each country under
study, we find a large number of institutions that are neither included in
the Ranking Web of Universities list nor in the list of OARE institutions.
For these institutions, we generate unique institution IDs as follows.
First, we order the institutions in a given country alphabetically.
Second, we identify all instances of a given institution in the raw data.
For instance, a given institution can have multiple versions because of
abbreviations, use of different languages, or typos. Thereby, we also use
the city where an institution is located to identify different versions of a
given institution, manually assigning identical institution IDs in such
cases.
Moreover, we assign institution IDs to track the relative position of
an institution in the university ranking list. For a given country, a lower
institution ID reflects a better rank. The rank variable, in addition, re-
flects the absolute worldwide position of the institution in the Ranking
Web of World Universities. This ranking provides information on the
performance of 22,123 research institutions worldwide on the basis of
the web presence as well as the impact of institutions. The former as-
pect is particularly noteworthy as web presence provides also a proxy
for the technical expertise needed to set up online access to journals.
Finally, we assign city IDs to construct distance and population
variables. To give an example, we identify 74 cities in Nigeria with a
population of more than 100,000 inhabitants (pop variable) using the
World Population Review (2017), listing population numbers for each
city in each country of the world. In addition, we identify 64 cities from
our Nigeria sample with fewer than 100,000 inhabitants. We assign city
IDs 1 to 138 to the Nigerian cities, where a lower number denotes a
larger population. As a further control, the variable distance_1 was
created by using Google maps and by computing the distance in km
from the city in which an institution is located to the largest domestic
city, as suggested by the first itinerary option by car. In addition, we
create a distance_2 variable indicating the distance of an institution's
location to the next domestic 1 million city.
3.4. Definition of variables
Table 1 provides an overview of the variables under study and
summary statistics at the institution-discipline-quarter level.
20
Vari-
ables can be grouped into six categories: dependent variable, countries,
main variable of interest, article characteristics, institutional char-
acteristics and city characteristics.
15
See Section 4 on the methodology. See also Section 3.4.2 on the definition
of our treatment variable.
16
Note that due to the complexity of the manual string matching process we
restrict our calculations to include multiple authored papers with up to 11
authors. This restriction does not reduce our sample by too much because 94%
of all papers have 11 or less authors.
17
Appendix 1 provides summary statistics for the sample of single local au-
thors.
18
We take the mean for the continuous variables, the max for the binary
variables and the sum for the publication variable in performing the collapse
command.
19
In total, 163 institutions in Nigeria, 96 in Peru, 82 in Kenya, 62 in Ecuador
and 56 in Bolivia. Please note that for all institutions that are not listed, we still
have the information that their rank is above 25,000. Since we use broadly
defined rank categories, we can therefore include rank information for all in-
stitutions under study.
20
Appendix 2 provides summary statistics by country group.
F. Mueller-Langer, et al. Research Policy 49 (2020) 103886
5
3.4.1. Dependent variables
Our dependent variable, y
s,t,r
, indicates the number of publications
of institution sin quarter tin discipline r. Recall from Section 3.3 that
we consider journal articles from both single local authors as well as
multiple local authors to create our dependent variable. In addition,
as reported in Table 1, the maximum number of publications per in-
stitution/quarter/discipline is 93.82. Hence, our dependent variable is
not a count variable (and we thus refrain from running a Poisson
model).
21
Following McCabe and Snyder (2015),Fig. 2 illustrates pat-
terns in publication output over time, i.e., from quarter 1 to quarter 50.
Publication output follows a steady upward trend, reaching a level
in the last quarter under study about 8% higher than in the base quarter
1. The results presented in Fig. 2 suggest that it is important to control
for secular trends in publication output.
22
As discussed in Section 4
(Methodology), we control for these secular trends in the regressions by
including binary variables for quarters at the discipline level.
3.4.2. Independent variables
Countries: We study 2,490 institutions from five countries of which
two are located in Sub-Saharan Africa (Kenya and Nigeria) and three in
South America (Bolivia, Ecuador, Peru). At the institution-discipline-
quarter level, 51.3% of our observations are from Sub-Saharan Africa.
Main variable of interest: (OARE) treated is our main variable of in-
terest. We construct this treatment variable by interacting three dummy
variables. First, OARE indicates whether papers are written by authors
affiliated with OARE institutions. We generate the OARE dummy by
using the institution IDs of all institutions that are part of UNEP's list of
OARE institutions. OARE (not reported in the table) takes on the value 1
if the respective institution of an article under study is an OARE in-
stitution and the value 0 otherwise. Second, the after dummy (not re-
ported in Table 1) accounts for the registration date (in quarters) of a
certain OARE institution. Its value is 1 if the article under study was
written by an author affiliated with an OARE institution after the in-
stitution joined the OARE program and 0 otherwise.
23
Recall that the
after dummy turns 1 in different quarters for different OARE institutions
as OARE institutions registered at different points in time. Third, we
generate an OARE research field dummy capturing whether a particular
article is within the top-50 OARE research areas or not. This allows us
to compare differences within institutions, i.e. differences between
disciplines that are core OARE research fields (e.g. environmental sci-
ence) versus non-OARE fields of research (e.g. economics).
Article characteristics: #co-authors USA (#co-authors EUR) indicates
the average number of co-authors from the US (Europe). #pages in-
dicates the average number of pages.
Institutional characteristics: Five variables indicate the rank of an
institution constructed from the Ranking Web of Universities (2014).
Rank1 represents the best institutions (rank ≤ 5,000) whereas Rank4
corresponds to institutions with the lowest reported ranks
(15,000 < rank≤25,000). Rank5 indicates that an institution is not
listed, which implies that its rank is above 25,000; these institutions are
Table 1
Summary Statistics.
mean sd min max N
Dependent variables
# publications 0.140 1.194 0 93.82 249,000
# references 20.60 21.27 0 293 249,000
# OARE references 4.293 7.222 0 135 249,000
Countries
Kenya 0.256 0.436 0 1 249,000
Nigeria 0.257 0.437 0 1 249,000
Bolivia 0.129 0.335 0 1 249,000
Ecuador 0.130 0.336 0 1 249,000
Peru 0.228 0.419 0 1 249,000
Main variable of interest
OARE treated (DDD) 0.007 0.0833 0 1 249,000
Article characteristics
# co-authors USA 0.472 1.329 0 37.25 249,000
# co-authors EUR 0.543 1.651 0 57 249,000
# pages 6.205 6.306 0 120 249,000
Institutional characteristics
Rank1: rank≤5,000 0.0201 0.140 0 1 249,000
Rank2: 5,000 < rank≤10,000 0.0161 0.126 0 1 249,000
Rank3: 10,000 < rank≤15,000 0.0181 0.133 0 1 249,000
Rank4: 15,000 < rank≤25,000 0.0253 0.157 0 1 249,000
Rank5: rank > 25,000 0.920 0.271 0 1 249,000
Rank, in 1,000 7.185 4.855 0.749 21.79 249,000
City characteristics
Distance from largest domestic city, in 100 km 3.254 3.613 0 20.64 249,000
Distance from closest dom. city with > 1
million inhabitants, in 100 km
1.991 3.216 0 20.64 249,000
Pop0: pop≤100, in 1,000 0.220 0.414 0 1 249,000
Pop1: 100 < pop≤500, in 1,000 0.106 0.308 0 1 249,000
Pop2: 500 < pop≤1,000, in 1,000 0.159 0.366 0 1 249,000
Pop3: 1,000 < pop≤5,000, in 1,000 0.335 0.472 0 1 249,000
Pop4: pop > 5,000, in 1,000 0.180 0.384 0 1 249,000
We use a balanced panel and take into account journal articles by both single and multiple local authors. Data is aggregated at the institution-
discipline-quarter level that constitutes our unit of observation.
21
The histogram of the number of publications at the institution-discipline-
quarter level is shown in Appendix 3.
22
Using a panel of citations to economics and management journals,
McCabe and Snyder (2015) explore whether online availability of journals in-
creases cites to scientific works. McCabe and Snyder's (2015) Fig. 2 illustrates
that citations follow a steady upward trend. Based on this finding, McCabe and
Snyder (2015) argue that it is important to account for these secular citation
trends in order for the online access variable of interest to be identified. Here,
we address the underlying identification problem in a similar fashion.
23
For non-OARE members after is set to 1 for all quarters after quarter 28
(launch of OARE).
F. Mueller-Langer, et al. Research Policy 49 (2020) 103886
6
the least productive in scientific output.
24
City characteristics: We use two different distance measures, in-
dicating (1) the distance in 100 km of a given city to the largest do-
mestic city (henceforth also distance_1) or (2) the distance in 100 km of
a given city to the next domestic 1 million city (henceforth also dis-
tance_2).
25
City population dummies indicate the number of inhabitants
of the city where an institution is located: Pop0 indicates cities with less
than 100,000 inhabitants whereas Pop4 indicates cities with more than
5,000,000 inhabitants.
4. Methodology
4.1. DDD using OLS regressions
In order to analyze the effect of the OARE initiative, we use a DDD
method for comparing the change in research output and input for re-
search fields in the treatment group (i.e. environmental sciences in
registered institutions after OARE registration) with the change in re-
search output for scientific fields in the control group (i.e. environ-
mental sciences in registered institutions before OARE registration,
non-environmental sciences in registered institutions and all research
fields in unregistered institutions) before and after a given institution
has registered (or not) with OARE. The intuition behind the DDD ap-
proach is the following. Within an OARE institution, only researchers
working on environmental issues can be impacted by free or reduced-
fee online access to environmental (OARE) journals after the institution
has registered with OARE. In contrast, other scientific fields within the
same institution (and OARE fields before OARE registration) will not
benefit from the OARE program. Exploring effects of online access
across scientific fields within a given institution mitigates concerns of
self-selection at the institutional level, for instance, because better/
more productive institutions might be more likely to register with
OARE. Following McCabe and Snyder (2015), we also include quarter-
discipline and institution-discipline fixed effects in the regressions to
account for the effects of institutions, disciplines and time on research
productivity.
The dependent variable, y
s,t,r
, is the number of published articles by
researchers from institution sin quarter tin research area r(henceforth,
discipline). We use the specification outlined in Eq. (1):
= + + + + +
= =
y fe fe x
k K t T
treated
with 1, , ; 1, ...
s t r t r t r s r s r s t r
k
kk s t r s t r
, , 0 1, , ,2, , ,3, , 4, , , , , ,
(1)
where fe
t,r
are quarter-discipline fixed effects (100 quarter-discipline
pairs in the balanced panel using the full sample); fe
s,r
are institution-
discipline fixed effects (4,890 institution-discipline pairs). Variable
treated
s,t,r
is our main variable of interest. It accounts for the fact that
institutions registered with the OARE initiative at different points in
time and that other disciplines than environmental sciences in a given
institution will not benefit from OARE. In other words, treated is 1 if an
institution is an OARE institution and if articles of affiliated researchers
are published in the OARE research discipline in a quarter after the
institution registered with OARE (and 0 otherwise). x
k,s,t,r
are kcontrol
variables (k=1, …, K).
26
ε
s,t,r
are unobservable effects assumed in-
dependent across s, t and r. When we refrain from including institution-
discipline fixed effects, we include institutional characteristics (such as
worldwide rank and the number of publications during which an in-
stitution published at least one article) and city characteristics (such as
distance to largest domestic city/next 1-million city and city popula-
tion).
We also explore the effect of OARE on publication input as mea-
sured by (a) the number of cited references and (b) the number of cited
Fig. 2. SECULAR TREND IN PUBLICATION OUTPUT Fig. 2 illustrates patterns in publication output over time, i.e., from quarter 1 to quarter 50. We use a balanced
panel and take into account journal articles by both single and multiple local authors. In order to be able to interpret the OLS results for the 49 quarter dummies
directly in terms of percentage changes compared to the reference quarter Q1, here we use the log of the number of publications of institution sin quarter tin
discipline ras dependent variable. Following McCabe and Snyder (2015,Fig. 2) the middle curve plots a set of quarter fixed effects from OLS regression using Q1 as
reference quarter. We use the xtreg command implemented in STATA. The underlying regression also includes binary variables for countries and quarter-discipline
pairs. Outside lines bound the 95% confidence interval based on robust standard errors clustered at the institution level. This figure illustrates that publication output
follows a steady upward trend, reaching a level in the last quarter under study about 8% higher than in the reference quarter Q1.
24
Using rank categories instead of the actual rank has the advantage of being
invariant to small variations in rank over time.
25
We do not have distance information for 206 institutions, as the respective
cities do not appear in Google maps. For these cities, we proxy the distance to
the largest domestic city by taking the average distance in the respective
country. We use the same approach for our alternative distance variable.
26
For instance, we include article characteristics such as the number of pages,
co-authors USA and co-authors EUR. We also include dummy variables in-
dicating the country where a given institution is located.
F. Mueller-Langer, et al. Research Policy 49 (2020) 103886
7
OARE references. For this purpose, we use again the specification
outlined in (1). The only difference is that y
s,t,r
now indicates (a) the
number of references in articles by researchers from institution sin
quarter tin research discipline r, and (b) the number of OARE refer-
ences in articles by researchers from institution sin quarter tin research
discipline r, respectively.
4.2. DDD using instrumental variables
A potential endogeneity problem arises if the unobservable variable
in the self-selection equation is correlated with the unobserved pro-
ductivity factors (see Section 3.2 above). There are two such un-
observable factors that we need to deal with. First, only well-informed
institutions can join the OARE initiative, especially since institutions
that had registered with earlier Research4Life initiatives (i.e., HINARI
and AGORA) may be better informed about the benefits of joining
OARE. This could lead to unobserved endogenous information pro-
blems that could explain the low OARE adoption rate. Since registration
with OARE is either costless (for the majority of the institutions of our
sample that belong to group A countries) or very cheap, direct mone-
tary cost are unlikely to be a factor for joining OARE. However, there
might be hidden administrative costs that could explain OARE mem-
bership. Secondly, unobserved endogenous problems resulting from an
insufficient ICT-infrastructure could also influence OARE membership.
Indeed, a sufficiently developed infrastructure is a prerequisite for on-
line access. We construct three instruments that can account for un-
observed informational and infrastructure effects at the institution
level. For each institution, we compute the average number of institu-
tions that have joined the OARE, the HINARI and the AGORA initiatives
in similar geographic areas of the country in which the institution is
located. More specifically, we constructed five distance categories ac-
cording to whether a given institution is located in a city with more
than one million inhabitants, less than 50 km away from such a city,
between 50 km and 250km, between 250 km and 750 km and more
than 750 km away from a city with more than one million inhabitants.
We also use the five categories of cities according to the size of the
population: less than 100,000, between 100,000 and 500,000, between
500,000 and 1 million, between 1 million and 5 million, more than 5
million. Then, we compute the average number of institutions that
joined the OARE, the AGORA and the HINARI initiatives – in similar
distance and population areas where a given institution is located – for
each quarter in the sample. From this we use the one-quarter lagged
values as instruments (respectively m_OARE, m_AGORA and m_HINARI).
These instruments should control for both different levels of awareness
about the existence of Research4Life before institutions join the OARE
program as well as infrastructure related issues that could lead to self-
selection into the OARE program. They should be uncorrelated with
individual unobserved productivity factors at the institution level and
thus can be used as valid instruments.
4.3. DDD using Bayesian MCMC estimation
We estimate the treatment effect using Bayesian estimation tech-
niques based on a data augmentation MCMC algorithm described in
Appendix 4 that can deal with endogeneity. There are two equations.
The first equation corresponds to the self-selection process and de-
termines the outcome of the binary treatment effect within a latent
variable framework. The second equation is the productivity Eq. (1).
We assume that the unobserved variables of both equations follow a
bivariate normal distribution with correlation coefficient ρ. The MCMC
algorithm simulates the latent variable of the first equation to generate
the endogenous binary treatment effect. The Bayesian approach ex-
plicitly deals with the correlation between the unobserved variables of
the two equations. If there are any unobserved variables that determine
whether an institution self-selects into the OARE program, the Bayesian
method accounts for its potential endogeneity on the estimation of the
treatment effect.
5. Empirical analysis
Our empirical analysis explores the effect of OARE on publication
output, applying triple difference regressions using OLS in Section 5.1.,
instrumental variables estimation (Section 5.2.) and Bayesian MCMC
estimation (Section 5.3.). In Section 5.4., we extend our analysis to the
effect of OARE on publication input, covering both OARE references as
well as total number of references. Section 5.5. provides additional
robustness checks.
5.1. The effect of OARE on publication output using OLS
We estimate the impact of OARE membership on scientific output by
using eight different specifications in Table 2. Specifications (1) to (8) use
OLS estimation.
27
Column (1) reports the OLS regression coefficients for
the basic model, including the treatment variable, country and quarter
dummy variables as well as dummy variables for 100 quarter-discipline
pairs. We add article characteristics in (2), institutional rank information
in (3), city population in (4), distance to the largest domestic city in (5)
and distance to the closest domestic city with more than 1 million in-
habitants in (6). In specification (7), we include institution-discipline
fixed-effects instead of country dummy variables and institutional and city
characteristics (rank, population and distance). In specification (8), we add
dummy variables indicating the numbers of quarters during which an
institution published at least one article.
We find a positive and robust OARE effect that is statistically sig-
nificant at the 1% level across all specifications.
28
The marginal OARE
effect ranges from +29.6% in specification (7) to +31.3% in specifi-
cation (8).
29
We also ran the regressions separately for Group A and
Group B countries (Appendix 5) and for each of the five countries
(Appendix 6). The OARE treatment effect is positive and statistically
significant for the subgroups. It is higher for institutions in Group A
countries (i.e. free access countries) than for institutions in group B
countries (i.e. reduced-fee countries).
Moving from column (1) to column (2), we consider the effects of
article characteristics on publication output. R-squared is similar (0.092
versus 0.085) and the OARE effect remains almost the same when we
include article characteristics in (2). R-squared increases from 0.085 to
0.107 while the OARE effect decreases only slightly when we add in-
stitutional rank information in (3).
30
We also find that lower-ranked institutions are less productive in
terms of publication output, since the coefficients associated with lower
ranks (5000 < rank≤10000, 10000 < rank≤15000 and rank25,000)
as compared to the best rank category rank≤5000 (reference category)
are negative and statistically significant at least at the 5% level across
columns (3) to (6).
31
27
We use the xtreg command in STATA. The institution-discipline-quarter
triplets constitute the unit of observation. Random institution-discipline effects
are included in columns (1) to (6).
28
All country dummy variables are negative. Recall that Nigeria is the re-
ference country. This suggests that Nigeria has the largest publication output.
29
We obtain these results by dividing the OARE treatment effect by the total
number of publications of an average institution over the full period. For in-
stance, for specification (7), we obtain 2.069/7.0 =0.296 and for specification
(8) 2.191/7.0 = 0.313, respectively.
30
Note that the Ranking Web of Universities that we use to create the rank
variable is mainly based on the assessment of the web presence of institutions,
e.g., it uses link analysis for quality evaluation. In this respect, an institution's
web performance provides a proxy for its technical expertise to set up online
access to journals.
31
In specification (8), the rank dummies capture productivity effects that are
not captured by the number of quarters with publications FE while in columns
(3) to (6) the rank dummies capture all productivity effects.
F. Mueller-Langer, et al. Research Policy 49 (2020) 103886
8
In addition, the distance to the largest domestic city has a negative
impact on output that is statistically significant at the 5% level. The
distance from the closest domestic city with more than 1 million in-
habitants has a negative impact on output that is not statistically sig-
nificant. Moreover, the distance variable also proxies the importance of
a sufficient ICT-infrastructure for online access to scientific content.
This suggests that institutions further away from a metropolitan region
may simply benefit less from free or reduced fee online access due to
insufficient ICT.
To mitigate concerns about outliers and about the large number of
quarters during which an institution did not publish any article, we
construct an alternative, binary dependent variable y. It is equal to 0
when an institution does not publish any article in a given quarter and
scientific field. It is equal to 1 for any positive article contribution of
affiliated authors. We run the same regressions as reported in Table 2
using this alternative, binary dependent variable. Results are reported
in Table 3. Coefficients can be directly interpreted as percentage effects.
The results reported in Table 3 suggest that OARE has a positive and
statistically significant effect on the probability of any positive article
contribution of affiliated authors. The marginal OARE effect ranges
from +15.8% in column (8) to +22.4% in column (1). These results
provide additional empirical support for a robust effect of OARE on
Table 2
Effect of OARE on Publication Output.
(1) (2) (3) (4) (5) (6) (7) (8)
Model: Base + Article info +Rank +Pop. +Dist._1 +Dist._2 +Inst.-Disc. FE # of quarters with
publications FE
Dependent variable: w w w w w w w w
OARE treated (DDD) 2.110*** 2.100*** 2.093*** 2.093*** 2.093*** 2.093*** 2.069*** 2.191***
(0.401) (0.401) (0.399) (0.399) (0.399) (0.399) (0.394) (0.0216)
# pages 0.00512* 0.00441 0.00431 0.00433 0.00434 0.00671* 0.000611
(0.00307) (0.00303) (0.00303) (0.00303) (0.00303) (0.00382) (0.000391)
# co-authors USA 0.00402 0.00376 0.00377 0.00379 0.00375 0.00137 −0.0069***
(0.0105) (0.0103) (0.0103) (0.0103) (0.0103) (0.0115) (0.00145)
# co-authors EUR 0.0431 0.0429 0.0428 0.0429 0.0429 0.0493 0.00279**
(0.0319) (0.0320) (0.0320) (0.0320) (0.0320) (0.0361) (0.00116)
Rank2: 5,000 < rank≤10,000 −0.819** −0.818** −0.822** −0.814** 0.00778
(0.395) (0.396) (0.396) (0.396) (0.0182)
Rank3: 10,000 < rank≤15,000 −0.907*** −0.888*** −0.897*** −0.888*** 0.0413**
(0.343) (0.342) (0.341) (0.342) (0.0180)
Rank4: 15,000 < rank≤25,000 −0.00279 0.00136 0.0108 0.00991 0.213***
(0.537) (0.533) (0.534) (0.533) (0.0165)
Rank5: rank > 25,000 −0.984*** −0.999*** −0.997*** −0.994*** 0.0859***
(0.325) (0.327) (0.327) (0.327) (0.0130)
Pop1: 100 < pop≤500, in 1,000 −0.0565 −0.0203 −0.0361 −0.0768***
(0.0488) (0.0506) (0.0532) (0.00965)
Pop2: 500 < pop≤1,000, in 1,000 0.0579 0.0973 0.0734 −0.0485***
(0.0667) (0.0730) (0.0718) (0.00979)
Pop3: 1,000 < pop≤5,000, in 1,000 0.0605 0.0356 0.0393 −0.0236***
(0.0630) (0.0584) (0.0637) (0.00913)
Pop4: pop > 5,000, in 1,000 0.00191 −0.0987 −0.0415 −0.0508***
(0.0364) (0.0634) (0.0413) (0.00931)
Distance from largest domestic city, in 100 km −0.0173**
(0.00837)
Distance from closest dom. city > 1 mill.
inhab., in 100 km
−0.00966 0.00113
(0.00619) (0.000807)
Constant 0.333*** 0.262*** 1.158*** 1.150*** 1.237*** 1.160*** 0.0451 −0.0501**
(0.0601) (0.0623) (0.319) (0.321) (0.322) (0.320) (0.0302) (0.0230)
Quarter dummies YES YES YES YES YES YES YES YES
Country dummies YES YES YES YES YES YES NO NO
Quarter-discipline dummies YES YES YES YES YES YES YES YES
Institution-discipline FE NO NO NO NO NO NO YES NO
# quarters with publications FE NO NO NO NO NO NO NO YES
Observations 249,000 249,000 249,000 249,000 249,000 249,000 249,000 249,000
R-squared, overall 0.0919 0.0846 0.1071 0.1079 0.1089 0.1081 0.0761 0.524
Number of Inst_Discipline 4,980 4,980 4,980 4,980 4,980 4,980 4,980 4,980
Number of Inst 2,490 2,490 2,490 2,490 2,490 2,490 2,490 2,490
We use a balanced panel and take into account journal articles by both single and multiple local authors. Results on the impact of OARE membership (treated) on
publication output of research institutions in five developing countries (Bolivia, Ecuador, Kenya, Nigeria, Peru) from OLS DDD. We use the xtreg command in STATA.
OLS regression coefficients reported. The institution-discipline-quarter triplets constitute the unit of observation. Period under study: 1st quarter 2000 to 2nd quarter
2012. Reference country is Nigeria. Reference quarter is 36. Reference rank is rank≤5000. Reference population is pop≤100. Random institution-discipline effects
are included in columns (1) to (6). Robust standard errors clustered at the institutional level. Note that serial correlation is not an issue in our balanced panel because
the large number of periods with 0 publications breaks any time correlation for any given institution. *p< 0.1, **p< 0.05, ***p< 0.01.
F. Mueller-Langer, et al. Research Policy 49 (2020) 103886
9
publication output.
5.2. The effect of OARE using IV estimation
In this section, we explicitly deal with endogeneity of unobserved
factors in the productivity equation. We first explain the probability to
join the OARE initiative and then use the predicted value of the treat-
ment effect in the main productivity equation. In the first stage of the IV
procedure, we explain the probability that an institution joins the OARE
initiative using quarter, country and institutions fixed effects as well as
the three instruments m_OARE, m_HINARI and m_AGORA using a linear
probability model estimated by ordinary least squares regression. In the
second stage of the IV procedure, we explain the number of publications
by the predicted treatment effect obtained from the first stage in ad-
dition to the other control variables used before.
Column (1) in Table 4 gives the estimation results of the IV proce-
dure using institution-discipline fixed effects in addition to quarter and
quarter-discipline fixed effects. The IV estimate (2.5) is slightly higher
Table 3
Effect of OARE on Publication Output (Linear Probability Model).
(1) (2) (3) (4) (5) (6) (7) (8)
Model: Base + Article info +Rank +Population +Distance_1 +Distance_2 +Inst.-Disc. FE +# quarters with
publications FE
Dependent variable: y y y y y y y y
OARE treated (DDD) 0.224*** 0.224*** 0.216*** 0.216*** 0.216*** 0.216*** 0.205*** 0.158***
(0.0219) (0.0219) (0.0224) (0.0224) (0.0224) (0.0224) (0.0230) (0.00502)
# pages 0.00107*** 0.000629*** 0.000587*** 0.000595*** 0.000604*** −0.000529*** −2.13e-05
(0.000154) (0.000165) (0.000166) (0.000166) (0.000166) (0.000144) (9.08e-05)
# co-authors USA 0.00274** 0.00262** 0.00257** 0.00258** 0.00255** −0.000864 3.50e-06
(0.00134) (0.00104) (0.00103) (0.00104) (0.00104) (0.000763) (0.000337)
# co-authors EUR −0.00163*** −0.00177*** −0.00176*** −0.00174*** −0.00174*** −0.00203*** −0.000146
(0.000484) (0.000546) (0.000549) (0.000549) (0.000547) (0.000739) (0.000271)
Rank2: −0.200*** −0.197*** −0.198*** −0.196*** 0.00140
5,000 < rank≤10,000 (0.0506) (0.0505) (0.0506) (0.0506) (0.00423)
Rank3: −0.219*** −0.213*** −0.214*** −0.213*** 0.00286
10,000 < rank≤15,000 (0.0486) (0.0484) (0.0484) (0.0485) (0.00417)
Rank4: −0.101* −0.0983* −0.0969* −0.0961* 0.00228
15,000 < rank≤25,000 (0.0555) (0.0554) (0.0553) (0.0554) (0.00383)
Rank5: rank > 25,000 −0.276*** −0.277*** −0.276*** −0.275*** 0.0121***
(0.0417) (0.0416) (0.0417) (0.0417) (0.00303)
Pop1: −0.000357 0.00482 0.00498 −0.000831
100 < pop≤500, in 1,000 (0.0100) (0.0107) (0.0109) (0.00224)
Pop2: 0.0162 0.0218* 0.0202* −0.000723
500 < pop≤1,000, in 1,000 (0.0106) (0.0115) (0.0113) (0.00228)
Pop3: 0.0131 0.00952 0.00755 −0.000576
1,000 < pop≤5,000, in 1,000 (0.00858) (0.00849) (0.00876) (0.00212)
Pop4: pop > 5,000, in 1,000 0.0123 −0.00205 0.000996 −0.000473
(0.00825) (0.0106) (0.00951) (0.00217)
Distance from largest domestic −0.00247**
city, in 100 km (0.00117)
Distance from closest dom. city −0.00252** 1.78e-05
with > 1 mill. inhab., in 100 km (0.00115) (0.000188)
Constant 0.127*** 0.117*** 0.377*** 0.370*** 0.382*** 0.373*** 0.0567*** −0.0129**
(0.00891) (0.00878) (0.0423) (0.0422) (0.0428) (0.0423) (0.00585) (0.00534)
Quarter dummies YES YES YES YES YES YES YES YES
Country dummies YES YES YES YES YES YES NO NO
Quarter-discipline dummies YES YES YES YES YES YES YES YES
Institution-discipline FE NO NO NO NO NO NO YES NO
# quarters with publications FE NO NO NO NO NO NO NO YES
Observations 249,000 249,000 249,000 249,000 249,000 249,000 249,000 249,000
R-squared, overall 0.0558 0.0583 0.0928 0.0934 0.0939 0.0937 0.005 0.396
Number of Inst_Discipline 4,980 4,980 4,980 4,980 4,980 4,980 4,980 4,980
Number of Inst 2,490 2,490 2,490 2,490 2,490 2,490 2,490 2,490
Dependent variable yis a binary variable equal 1 if an institution made any article contribution in a given quarter and discipline. We use a balanced panel and take
into account journal articles by both single and multiple local authors. Results on the impact of OARE membership (treated) on any publication output of research
institutions in five developing countries (Bolivia, Ecuador, Kenya, Nigeria, Peru). We explore a linear probability model. We use the xtreg command in STATA. OLS
regression coefficients reported. The institution-discipline-quarter triplets constitute the unit of observation. Period under study: 1st quarter 2000 to 2nd quarter
2012. Reference country is Nigeria. Reference quarter is 36. Reference rank is rank≤5000. Reference population is pop≤100. Random institution-discipline effects
are included in columns (1) to (6). Robust standard errors clustered at the institutional level. Note that serial correlation is not an issue in our balanced panel because
the large number of periods with 0 publications breaks any time correlation for any given institution.
*p< 0.1, **p< 0.05, ***p< 0.01.
F. Mueller-Langer, et al. Research Policy 49 (2020) 103886
10
than the OLS estimate (2.1). However, the standard error is quite large.
Column (2) reports the estimation results of the linear probability
model where the dependent variable is y. Coefficients can be directly
interpreted as percentage effects. Compared to the linear probability
model effect estimated by OLS (21%), the IV procedure yields a higher
treatment effect of 74%. Again, the standard error is quite high.
5.3. The effect of OARE using Bayesian MCMC estimation
In this section, we explicitly model correlation between the un-
observed variable in the self-selection equation and the unobserved
variable in the productivity equation by using a Bayesian MCMC esti-
mation method. Here, we simultaneously estimate parameters of a
system of two equations using Bayesian inference: parameters of the
self-selection equation are jointly estimated with parameters of the
productivity equation assuming a normal bivariate distribution of the
unobserved variables of the two equations. Inference from the posterior
distribution of the parameters is carried out using data augmentation
and an MCMC algorithm. Contrary to the IV approach, the Bayesian
approach is an exact likelihood procedure.
32
Results from Table 5 were obtained from a random sample of 300
institutions. Columns (1) and (2) report results from OLS and MCMC
regressions with was dependent variable, respectively. Columns (3) and
(4) report results from a linear probability model with yas dependent
variable. The sample was selected to match the treatment effect esti-
mated on the full sample by OLS as reported in Tables 2 and 3. More
specifically, the random sample was selected so that the treatment ef-
fect in column (1) of Table 5 (2.06 in the second stage equation using
OLS) was matched with the estimated treatment effect of specification
(7) of Table 2 (2.07). In addition, the random sample also matched the
treatment effect using yas a dependent variable in column (3) of
Table 5 (0.21 in the second stage of the linear probability model) with
the estimated treatment effect of column (7) of Table 3 (0.205). In all
specifications reported in Table 5, we include a set of binary variables
corresponding to the number of quarters during which an institution
published at least one article as institution fixed effects, in addition to
country, rank, population and distances variables.
33
Across columns,
the Bayesian MCMC estimate of the treatment effect is slightly above
the OLS estimate with a small standard error, i.e., 2.099 > 2.06 and
0.224 > 0.210. The correlation between unobservable variables of the
selection and observation equations is slightly negative and significant,
respectively −0.056 with was the dependent variable and −0.075
with yas the dependent variable. Unobserved variables in the self-se-
lection equation include the hidden (administrative and informational)
costs of joining the initiative, while unobserved variables in the main
equation include hidden productivity factors. A negative correlation
between the unobserved variables corresponds to a negative correlation
between the hidden costs of joining the OARE initiative and the un-
observed productivity variables at the institution level.
5.4. Extensions
5.4.1. OARE effect by number of quarters with publications
We also ran the regressions separately for institutions with three
different levels of productivity, i.e., institutions that published in at
least xquarters in at least one of the two disciplines with x< 2,
2 ≤ x≤ 30, and x> 30, respectively. Results are reported in Table 6.
Specification (1) reports results for institutions with publications in less
than 2 quarters. Specification (2) reports results for institutions with
publications in between 2 and 30 quarters. Finally, specification (3)
reports results for the most productive institutions that publish in more
than 30 quarters of our sample (with a total of 50 quarters).
The marginal OARE effect ranges from +0.4% for institutions with
a low level of productivity as reported in column (1) to +43.4% for
institutions with a high level of productivity as reported in column (3).
It is statistically significant at the 1% level for institutions with inter-
mediate and high levels of productivity as reported in columns (2) and
(3). It is statistically significant at the 5% level for low-productivity
institutions as reported in column (1). Notably, these findings suggest
that productive institutions benefit more from OARE registration and
that the least productive institutions barely benefit from it.
5.4.2. The effect of OARE on publication input
In addition to the impact of the OARE initiative on publication
output, we investigate the effect of OARE on publication input.
Thereby, we address the cumulative character of the knowledge pro-
duction process and revert to the fact that researchers need access to
academic journals in order to be able to create their own research. Our
Table 4
OARE Effect using IV.
(1) (2)
Model: IV LPM IV
First Stage
Dependent variable: OARE treated OARE treated
m_OARE 0.612*** 0.612***
(0.0106) (0.0106)
m_AGORA 0.0145 0.0145
(0.0127) (0.0127)
m_HINARI −0.0162 −0.0162
(0.0124) (0.0124)
Constant 0.000606 0.000606
(0.000980) (0.000980)
Observations 249,000 249,000
R-squared 0.042 0.042
Number of Inst_Discipline 4,980 4,980
Quarter FE YES YES
Quarter-discipline FE YES YES
Inst-discipline FE YES YES
Second Stage
Dependent variable: w y
OARE treated (DDD) 2.455*** 0.743***
(0.541) (0.139)
Constant 0.126*** 0.0602***
(0.00802) (0.00324)
Observations 249,000 249,000
R-squared 0.016 0.018
Number of Inst_Discipline 4,980 4,980
Quarter FE YES YES
Quarter-discipline FE YES YES
Inst-discipline FE YES YES
We use a balanced panel and take into account journal articles by both single
and multiple local authors. Regressions based on Specification (7) of Table 2.
The institution-discipline-quarter triplets constitute the unit of observation.
Period under study: 1st quarter 2000 to 2nd quarter 2012. Reference country is
Nigeria. Reference quarter is 36. Reference rank is rank≤5000. Reference po-
pulation is pop≤100. Robust standard errors clustered at the institutional level.
Included but not reported: # pages, # co-authors USA, # co-authors EUR. ***
p< 0.01, ** p< 0.05, * p< 0.1.
32
We use a non-informative prior distribution of the parameters. The main
drawback of the MCMC procedure is that it is extremely computer intensive.
The MCMC could not run on the full set of 2,490 institutions with all fixed
effects.
33
Using the number of quarters as additional fixed effects allows us to run the
program on a relatively large sample of institutions. The MCMC algorithm was
run for 1,000 iterations after a warmup period of 100 iterations starting at the
OLS estimate and an initial value of ρof −0.15. Note that the coefficients in the
first stage of Table 5 are different from those in Table 4. Indeed, the first stage
of the IV procedure is a linear probability model, while the first equation of the
MCMC procedure is a standard probit model.
F. Mueller-Langer, et al. Research Policy 49 (2020) 103886
11
a priori belief based on the results originally formulated by
Davis (2011) is that free or reduced fee online access to academic
journals in environmental sciences should have an effect on access and
hence the number of references used to create one's own research. In
this regard, we look at both the overall number of references as well the
number of OARE references, i.e., references to articles published in
OARE journals.
The results on the impact of OARE on the number of references are
reported in Table 7. We find a positive effect of membership to OARE
on the number of references across all specifications. It is statistically
significant at the 1% level across all specifications. Interestingly, next to
OARE membership international collaborations with researchers from
the US (and to a smaller degree also from Europe) have a positive and
statistically significant effect on the number of references. This result
suggests that collaborations with researchers from the US (and Europe)
have a positive effect on developing-country researchers in terms of
access to scientific journals.
The results on the impact of OARE on the number of OARE
Table 5
OARE Effect using MCMC.
(1) (2) (3) (4)
Model: OLS MCMC OLS LPM MCMC LPM
First Stage
Dependent variable: OARE treated OARE treated OARE treated OARE treated
Average # of institutions 0.810*** 6.371*** 0.810*** 6.8221***
that joined OARE (t-1) (0.031) (1.054) (0.031) (1.1009)
Average # of institutions 0.256*** 4.011* 0.256*** 3.0013
that joined AGORA (t-1) (0.041) (2.060) (0.041) (1.9066)
Average # of institutions −0.3007*** −6.43*** −0.3007*** −6.247***
that joined HINARI (t-1) (0.042) (2.252) (0.042) (1.8116)
Constant −3.909*** −3.664***
(0.623) (0.5529)
Second Stage
Dependent variable: w w y y
OARE treated (DDD) 2.06*** 2.099*** 0.2103*** 0.2237***
(0.051) (0.054) (0.0124) (0.0131)
# pages −0.002*** −0.002*** 0.0009*** 0.0009***
(0.0007) (0.0007) (0.0001) (0.0001)
# co-authors USA 0.0077* 0.0080* 0.0017 0.0016
(0.0044) (0.0041) (0.0010) (0.0011)
# co-authors EUR −0.0008 −0.0008 −0.0004 −0.0004
(0.0032) (0.0032) (0.0007) (0.0007)
Rank2: 5,000 < rank≤10,000 −0.775*** −0.777*** 0.0720*** 0.0720***
(0.0528) (0.0511) (0.0127) (0.0128)
Rank3: 10,000 < rank≤15,000 −0.736*** −0.736*** 0.0049 0.0055
(0.0466) (0.0447) (0.0113) (0.0117)
Rank4: 15,000 < rank≤25,000 −0.726*** −0.727*** 0.0217** 0.0222**
(0.0447) (0.0445) (0.0108) (0.0111)
Rank5: rank > 25,000 −0.718*** −0.717*** 0.0138 0.0150*
(0.0347) (0.0336) (0.0085) (0.0088)
Pop1: 100 < pop≤500, in 1,000 −0.022 −0.022 −0.005 −0.005
(0.0233) (0.0218) (0.0056) (0.0057)
Pop2: 500 < pop≤1,000, in 0.0124 0.0123 −0.012** −0.012**
1,000 (0.0244) (0.0228) (0.0059) (0.0058)
Pop3: 1,000 < pop≤5,000, in 0.0685*** 0.0678*** −0.004 −0.004
1,000 (0.0201) (0.0193) (0.0049) (0.0049)
Pop4: pop > 5,000, in 1,000 −0.0217 −0.022 −0.007 −0.007
(0.0222) (0.0213) (0.0053) (0.0053)
Distance from closest dom. city 0.0033* 0.0032* −0.0002 −0.0002
with > 1 mill. inhab., in 100 km (0.0020) (0.0019) (0.0004) (0.0005)
Constant 0.127*** 0.675*** 0.0208 0.0196
(0.0158) (0.050) (0.0127) (0.0128)
ρ −0.056* −0.075**
(0.0317) (0.0340)
Quarter dummies YES YES YES YES
Country dummies YES YES YES YES
Quarter-discipline FE YES YES YES YES
# quarters with publications FE YES YES YES YES
Observations 30,000 30,000 30,000 30,000
Number of Inst_Discipline 600 600 600 600
Number of Inst 300 300 300 300
We use a balanced panel and take into account journal articles by both single and multiple local authors. Results obtained from a random sample of 300 institutions.
The institution-discipline-quarter triplets constitute the unit of observation. Columns (1) and (2) report results from OLS and MCMC regressions with was dependent
variable, respectively. Columns (3) and (4) report results from a linear probability model with yas dependent variable. First equation of the MCMC procedure is a
standard Probit model. Robust standard errors clustered at the institutional level. *p< 0.1, **p< 0.05, ***p< 0.01.
F. Mueller-Langer, et al. Research Policy 49 (2020) 103886
12
references are reported in Table 8. We find a positive effect of mem-
bership to OARE on the number of OARE references across all specifi-
cations. It is statistically significant at the 1% level across all specifi-
cations.
5.5. Robustness
5.5.1. Pre- and post-OARE trends
There may be concerns that the post-OARE effect that we obtain is
confounded with a pre-OARE trend which may undermine the inter-
pretation of the OARE effect as a treatment effect.
34
To mitigate these
concerns we implement a specification similar to specification (7) of
Table 2 following the approach taken by Furman and Stern (2011,
Fig. 2). Therein, we replace the OARE treated dummy with the dummy
variables for each quarter preceding and following the quarter when an
institution registered with OARE. From this new specification, we ob-
tain Fig. 3.
Fig. 3 plots quarter-by-quarter pre-treatment and post-treatments
effects on publication output. The data represents each of the estimated
marginal pre- and post-treatment quarter effects on publication output.
All marginal effects are computed relative to the treatment quarter
plus/minus one quarter. Outside lines bound the 95% confidence in-
terval based on robust standard errors clustered at the institution level.
Fig. 3 does not suggest that publication output follows a clear upward
trend in the 28 quarters before the OARE treatment.
5.5.2. Articles with only one local author
In our basic model, we use a balanced panel including both single-
and multiple-local-authored journal articles to examine the OARE ef-
fect. As discussed in Section 3.3, dividing authorship shares across
different institutions is not a trivial exercise. For robustness, we
therefore run the same regressions as reported in Table 2 for the sub-
sample of single-local-author articles, i.e. articles for which we observe
only one local author who may or may not be affiliated with an OARE
member institution. As before, we use a balanced panel. We create the
single-local-authored dataset by dropping 18,955 articles from the
sample for which we have at least two local authors. Results are re-
ported in Table 9.
The marginal OARE effect is positive and statistically significant at
the 1% level across all specifications. It ranges from +23.5% in column
(7) to +24.3% in column (1).
35
These results provide additional em-
pirical evidence for a robust OARE effect.
5.5.3. Effect of outliers
We address possible concerns that OLS results without log-trans-
formed dependent variable are potentially sensitive to extreme outliers.
Following Williams (2016), we delete observations that have at least
one of the following characteristics: (a) standardized residuals' values
greater than 3, (b) leverage greater than 2k/nwhere kis the number of
independent variables in the regression and nis the number of ob-
servations, (c) Cook's Distance measure values greater than 4/n. In
total, 4,709 observations are deleted. 1,355 observations are deleted
under (a), 13 under (b) and 3,341 under (c). Results are reported in
Table 10.
The OLS OARE coefficient is statistically significant at the 1% level
across all columns ranging from 1.48 in column (7) to 1.53 in column
(2). These results provide additional evidence for a robust OARE ef-
fect.
36
5.5.4. Unbalanced panel
In order to address possible concerns that there are many observa-
tions with w= 0 in the balanced-panel analysis, we use an unbalanced
panel and take into account journal articles by both single and multiple
local authors. Results are reported in Table 11.
The OLS OARE coefficient is statistically significant at the 1% level
across all columns ranging from 2.94 in column (7) to 4.95 in column
(1). These results provide additional evidence for a robust OARE effect.
6. Conclusion
We have analyzed the effect of free and reduced-fee online access to
the environmental science literature via the OARE initiative on scien-
tific productivity in Bolivia, Ecuador, Kenya, Nigeria and Peru. We
provide empirical support for a positive marginal OARE effect of 29.6%.
Given an average publication output by institution, discipline and
quarter of 0.140 (see Table 1), this result suggests that, due to the OARE
program, 2.07 extra articles were produced for an average institution
publishing a total of 7.0 articles over the whole observation period.
Looking at a 95% confidence interval this suggests that the OARE
program adds between 1.3 and 2.8 extra articles for an average in-
stitution over the whole period of 50 quarters. The marginal OARE
Table 6
OARE Effect by the Number of Quarters with Publications.
(1) (2) (3)
# quarters with publication: < 2 quarters 2 ≤ quarters
≤30
> 30 quarters
Model: +Inst.-Disc. FE +Inst.-Disc. FE +Inst.-Disc. FE
Dependent variable: w w w
OARE treated (DDD) 0.0246** 0.445*** 3.036***
(0.0118) (0.110) (0.969)
# pages [omitted] 0.00116 −0.0103
(0.00176) (0.00885)
# co-authors USA [omitted] −0.00795 0.0258
(0.00554) (0.0226)
# co-authors EUR [omitted] 0.00780*** 0.0939**
(0.00302) (0.0430)
Constant 0.00473 0.0643*** 3.182***
(0.00380) (0.0177) (0.393)
Quarter dummies YES YES YES
Country dummies NO NO NO
Quarter-discipline dummies YES YES YES
Institution-discipline FE YES YES YES
Observations 130,100 109,800 9,100
R-squared, overall 0.0009 0.0451 0.1775
Number of Inst_Discipline 2,602 2,196 182
Number of Inst 1,301 1,098 91
We use a balanced panel and take into account journal articles by both single
and multiple local authors. OLS results on the impact of OARE membership
(treated) on publication output of research institutions with three different le-
vels of productivity, i.e., institutions that published in at least xquarters in at
least one of the two disciplines whereas x< 2, 2 ≤ x≤ 30, and x> 30, re-
spectively. We use the xtreg command in STATA. OLS regression coefficients
reported. The institution-discipline-quarter triplets constitute the unit of ob-
servation. Period under study: 1st quarter 2000 to 2nd quarter 2012. Reference
country is Nigeria. Reference quarter is 36. Reference rank is rank≤5000.
Reference population is pop≤100. Robust standard errors clustered at the in-
stitutional level. Note that serial correlation is not an issue in our balanced
panel because the large number of periods with 0 publications breaks any time
correlation for any given institution. *p< 0.1, **p< 0.05, ***p< 0.01.
34
We thank an anonymous referee for this comment.
35
From Appendix 1 we know that the mean number of publications for the
single local author sample is 0.112. Hence, an average institution publishes 5.6
single-local authored articles over the observation period. Based on this, we
obtain 1.315/5.6 = 0.2348 for column (7) of Table 9 and 1.362/5.6 = 0.2432
for column (1) of Table 9, respectively.
36
Note that among the deleted observations are those institution/quarter/
discipline pairs with w> 80 publications in the OARE field. Recall from Table 1
that the mean of wis 0.14. Hence, the exclusion of extreme outliers in terms of
publication output in the OARE field explains that the OARE effect is lower than
in the case where outliers are not excluded.
F. Mueller-Langer, et al. Research Policy 49 (2020) 103886
13
effect is also positive and statistically significant when we run the re-
gressions separately for Group A (free access) and Group B (reduced-fee
access) countries, revealing at least 2.45 extra articles for institutions in
group A countries and 1.21 extra articles for institutions in group B
countries, respectively.
37
In addition, a robustness check analyzing a
balanced panel with single local authors yields qualitatively similar
results, i.e., we provide evidence for a statistically significant OARE
effect of +23.5% with confidence interval (15.1%, 31.9%) using the
most conservative specification. Overall, our results provide empirical
support for the hypothesis that free online access to journals promotes
research in developing countries. Moreover, we analyze the impact of
OARE on the ability of researchers in the developing world to access
academic knowledge. In line with Davis (2011)’s results we find a
strong impact of the OARE initiative on both the overall number of
references in articles to which an average institution contributed, i.e., a
+8.4% increase of the number of references due to OARE, as well as the
number of OARE references, i.e., a +14.5% increase in the number of
OARE references due to OARE. However, our result of a significant
OARE effect on publication output differs from Davis (2011) who finds
no TEEAL effect on publication output. Arguably, the larger number of
journals covered under the OARE initiative as compared to TEEAL
(11,500 vs. 200 journals), OARE's lower subscription cost (ranging from
free online access to a maximum of $ 1,000 per year vs. $ 5,000 for
TEEAL), and the different means of access (online access for OARE and
offline (CD-ROM) for TEEAL) provide an explanation why we find a
substantial OARE effect on publication output while Davis (2011) finds
no TEEAL effect on publication output.
Nevertheless, we find that there is potential for improvement on two
grounds. First, we provide evidence that productive institutions benefit
more from OARE and that the least productive institutions barely benefit
from it. This result suggests that OARE increases the productivity differ-
ence between the most and least productive institutions. Under these
conditions, the least productive institutions are ceteris paribus less likely to
Table 7
Effect of OARE on Publication Input (References).
(1) (2) (3) (4) (5) (6) (7)
Model: Base + Article info +Rank +Population +Distance_1 +Distance_2 +Inst.-Disc. FE
Dependent variable: #References #References #References #References #References #References #References
OARE treated (DDD) 2.017*** 1.716*** 1.711*** 1.711*** 1.711*** 1.711*** 1.722***
(0.335) (0.295) (0.295) (0.295) (0.295) (0.295) (0.295)
# pages 1.774*** 1.773*** 1.774*** 1.773*** 1.773*** 1.754***
(0.0775) (0.0775) (0.0776) (0.0776) (0.0776) (0.0791)
# co-authors USA 0.672*** 0.672*** 0.672*** 0.672*** 0.672*** 0.660***
(0.231) (0.232) (0.232) (0.232) (0.232) (0.233)
# co-authors EUR 0.378* 0.378* 0.378* 0.378* 0.378* 0.372*
(0.195) (0.195) (0.195) (0.195) (0.195) (0.194)
Rank2: 5,000 < rank≤10,000 −1.249 −1.296 −1.269 −1.416
(1.672) (1.682) (1.683) (1.672)
Rank3: 10,000 < rank≤15,000 0.379 0.303 0.366 0.306
(2.085) (2.079) (2.084) (2.087)
Rank4: 15,000 < rank≤25,000 0.380 0.436 0.370 0.216
(1.523) (1.525) (1.516) (1.498)
Rank5: rank > 25,000 −3.539*** −3.427*** −3.443*** −3.540***
(1.138) (1.153) (1.159) (1.157)
Pop1: 100 < pop≤500, in 1,000 0.121 −0.133 −0.407
(0.720) (0.750) (0.767)
Pop2: 500 < pop≤1,000, in 1,000 0.266 −0.0109 −0.137
(0.683) (0.721) (0.706)
Pop3: 1,000 < pop≤5,000, in 1,000 −0.610 −0.435 −0.0620
(0.545) (0.550) (0.589)
Pop4: pop > 5,000, in 1,000 −0.406 0.299 0.718
(0.732) (0.885) (0.870)
Distance from largest domestic city, in 0.121
100 km (0.0896)
Distance from closest domestic city 0.250**
with > 1 mill. inhab., in 100 km (0.117)
Constant 27.63*** 11.63*** 14.67*** 14.68*** 14.08*** 14.41*** 8.933***
(0.462) (0.741) (1.430) (1.456) (1.511) (1.457) (0.503)
Quarter dummies YES YES YES YES YES YES YES
Country dummies YES YES YES YES YES YES NO
Quarter-discipline dummies YES YES YES YES YES YES YES
Institution-discipline FE NO NO NO NO NO NO YES
Observations 249,000 249,000 249,000 249,000 249,000 249,000 249,000
R-squared, overall 0.2884 0.6325 0.6339 0.6342 0.6344 0.6346 0.6283
Number of Inst_Discipline 4,980 4,980 4,980 4,980 4,980 4,980 4,980
Number of Inst 2,490 2,490 2,490 2,490 2,490 2,490 2,490
We use a balanced panel and take into account journal articles by both single and multiple local authors. Results on the impact of OARE membership (treated) on the
mean number of references (i.e., by institution, quarter and discipline) of research institutions in five developing countries (Bolivia, Ecuador, Kenya, Nigeria, Peru)
from OLS DDD estimation methods. We use the xtreg command in STATA. OLS regression coefficients reported. The institution-discipline-quarter triplets constitute
the unit of observation. Period under study: 1st quarter 2000 to 2nd quarter 2012. Reference country is Nigeria. Reference quarter is 36. Reference rank is
rank≤5000. Reference population is pop≤100. Random institution-discipline effects are included in columns (1) to (6). Robust standard errors clustered at the
institutional level. *p< 0.1, **p< 0.05, ***p< 0.01.
37
We estimate these numbers as follows. The average number of articles is
0.175 for group A and 0.0768 for group B countries. The OARE program in-
creases publication output by 28% in group A and by 32% in group B countries.
This adds up to 2.45 (1.21) extra articles for an average institution in group A
(B) countries.
F. Mueller-Langer, et al. Research Policy 49 (2020) 103886
14
catch up. Second, we find that not more than 5% of all eligible institutions
joined OARE after a period of more than 5 years. This finding reveals the
unused potential of the OARE initiative. Based on our results, policies
aimed at increasing the awareness of free online access initiatives in de-
veloping countries should therefore be encouraged.
As a broader policy implication, our study suggests that an open
access mandate or policy may promote scientific output – not only by
research institutions in developing countries. Extending on the link
between academic research and economic growth (see the literature
discussed in Section 2) our findings may hence point to direct economic
effects as a higher research output level stemming from OARE mem-
bership may result in new environmental innovations. A natural follow
up is to explore the question of whether OARE has increased the
number of patent applications using free or reduced-fee access
throughout the OARE program. In addition, it would be interesting to
investigate in more detail how (open) online access has changed the
way scientists do research and collaborate internationally.
Declaration of Competing Interest
The authors declare that they have no known competing financial
interests or personal relationships that could have appeared to influ-
ence the work reported in this paper.
Table 8
Effect of OARE on Publication Input (OARE References).
(1) (2) (3) (4) (5) (6) (7)
Model: Base + Article info +Rank +Population +Distance_1 +Distance_2 +Inst.-Disc. FE
Dependent variable: #OARE_References #OARE_References #OARE_References #OARE_References #OARE_References #OARE_References #OARE_References
OARE treated (DDD) 0.736*** 0.622*** 0.621*** 0.621*** 0.621*** 0.621*** 0.622***
(0.162) (0.154) (0.154) (0.154) (0.154) (0.154) (0.154)
# pages 0.425*** 0.425*** 0.425*** 0.425*** 0.425*** 0.423***
(0.0344) (0.0344) (0.0344) (0.0344) (0.0344) (0.0353)
# co-authors USA 0.297*** 0.297*** 0.298*** 0.298*** 0.298*** 0.295***
(0.110) (0.110) (0.110) (0.110) (0.110) (0.111)
# co-authors EUR 0.240** 0.240** 0.240** 0.240** 0.240** 0.238**
(0.116) (0.116) (0.116) (0.116) (0.116) (0.117)
Rank2: 5,000 < rank≤10,000 1.216 0.771 0.798 0.680
(0.865) (0.870) (0.876) (0.871)
Rank3: 10,000 < rank≤15,000 0.600 0.0687 0.130 0.0710
(0.706) (0.693) (0.700) (0.699)
Rank4: 15,000 < rank≤25,000 0.793 0.522 0.457 0.357
(0.767) (0.749) (0.738) (0.728)
Rank5: rank > 25,000 −0.228 −0.262 −0.278 −0.347
(0.565) (0.566) (0.572) (0.572)
Pop1: 100 < pop≤500, in 1,000 −0.763** −1.011** −1.159***
(0.369) (0.394) (0.405)
Pop2: 500 < pop≤1,000, in 1,000 −1.083*** −1.354*** −1.386***
(0.372) (0.389) (0.380)
Pop3: 1,000 < pop≤5,000, in 1,000 −1.090*** −0.919*** −0.678**
(0.298) (0.297) (0.312)
Pop4: pop > 5,000, in 1,000 −2.271*** −1.581*** −1.427***
(0.366) (0.436) (0.419)
Distance from largest domestic city, 0.119**
in 100 km (0.0510)
Distance from closest domestic city 0.188***
with > 1 mill. inhab., in 100 km (0.0682)
Constant 4.873*** 0.892*** 0.995 1.880*** 1.284* 1.672** 1.291***
(0.166) (0.333) (0.654) (0.670) (0.710) (0.676) (0.250)
Quarter dummies YES YES YES YES YES YES YES
Country dummies YES YES YES YES YES YES NO
Quarter-discipline dummies YES YES YES YES YES YES YES
Institution-discipline FE NO NO NO NO NO NO YES
Observations 249,000 249,000 249,000 249,000 249,000 249,000 249,000
R-squared, overall 0.1342 0.2907 0.2918 0.2995 0.3008 0.3017 0.2746
Number of Inst_Discipline 4,980 4,980 4,980 4,980 4,980 4,980 4,980
Number of Inst 2,490 2,490 2,490 2,490 2,490 2,490 2,490
We use a balanced panel and take into account journal articles by both single and multiple local authors. Results on the impact of OARE membership (treated) on the
mean number of OARE references (i.e., by institution, quarter and discipline) of research institutions in five developing countries (Bolivia, Ecuador, Kenya, Nigeria,
Peru) from OLS DDD estimation methods. We use the xtreg command in STATA. OLS regression coefficients reported. The institution-discipline-quarter triplets
constitute the unit of observation. Period under study: 1st quarter 2000 to 2nd quarter 2012. Reference country is Nigeria. Reference quarter is 36. Reference rank is
rank≤5000. Reference population is pop≤100. Random institution-discipline effects are included in columns (1) to (6). Robust standard errors clustered at the
institutional level. *p< 0.1, **p< 0.05, ***p< 0.01.
F. Mueller-Langer, et al. Research Policy 49 (2020) 103886
15
Fig. 3. PRE- AND POST-TREATMENT EFFECTS ON PUBLICATION OUTPUT Notes: Figure plots quarter-by-quarter pre-treatment and post-treatment effects on
publication output, computed from OLS triple difference regressions with dummy variables for each quarter preceding and following OARE treatment (along with
quarter fixed effects, institution-discipline fixed effects and article characteristics). The regressions reported in column (7) of Table 2 serve as the basis for the
regressions from which create the figure. In contrast to the regressions reported in column (7) of Table 2, however, we replace the OARE treated dummy with the
dummy variables for each quarter preceding and following treatment. The data represent each of the estimated marginal pre- and post-treatment quarter effects on
publication output. All marginal effects are computed relative to the treatment quarter plus/minus one quarter. Outside lines bound the 95% confidence interval
based on robust standard errors clustered at the institution level. Fig. 3 does not suggest that publication output follows a clear upward trend in the 28 quarters before
the OARE treatment.
Table 9
OARE Effect (SINGLE Local Authors ONLY).
(1) (2) (3) (4) (5) (6) (7)
VARIABLES OLS Base + Article info +Rank +Population +Dist._1 +Dist._2 +Inst.-Disc. FE
OARE treated (DDD) 1.362*** 1.347*** 1.339*** 1.339*** 1.339*** 1.339*** 1.315***
(0.249) (0.245) (0.244) (0.244) (0.244) (0.244) (0.240)
# pages 0.00679** 0.00598* 0.00591* 0.00594* 0.00594* 0.00995**
(0.00317) (0.00312) (0.00313) (0.00313) (0.00313) (0.00441)
# co-authors USA −0.00484 −0.00501 −0.00504 −0.00507 −0.00511 −0.0123
(0.0123) (0.0122) (0.0122) (0.0122) (0.0122) (0.0163)
# co-authors EUR 0.0562 0.0558 0.0557 0.0558 0.0558 0.0681
(0.0356) (0.0356) (0.0356) (0.0356) (0.0356) (0.0414)
Rank2: 5,000 < rank≤10,000 −0.638** −0.641** −0.642** −0.637**
(0.259) (0.259) (0.260) (0.260)
Rank3: 10,000 < rank≤15,000 −0.641*** −0.629*** −0.637*** −0.630***
(0.243) (0.242) (0.241) (0.242)
Rank4: 15,000 < rank≤25,000 −0.0311 −0.0331 −0.0291 −0.0263
(0.373) (0.369) (0.370) (0.370)
Rank5: rank > 25,000 −0.742*** −0.754*** −0.752*** −0.751***
(0.219) (0.221) (0.221) (0.221)
Pop1: 100 < pop≤500, in 1,000 −0.0483 −0.0259 −0.0333
(0.0396) (0.0413) (0.0425)
Pop2: 500 < pop≤1,000, in 1,000 0.0313 0.0576 0.0430
(0.0496) (0.0543) (0.0525)
Pop3: 1,000 < pop≤5,000, in 1,000 0.0319 0.0144 0.0139
(0.0466) (0.0439) (0.0468)
Pop4: pop > 5,000, in 1,000 −0.0124 −0.0800* −0.0486
(0.0300) (0.0458) (0.0334)
Distance from largest domestic −0.0115**
city, in 100 km (0.00575)
Distance from closest domestic city −0.00773*
with > 1 mill. inhab., in 100 km (0.00460)
Constant 0.259*** 0.166*** 0.840*** 0.842*** 0.899*** 0.851*** 0.0453
(0.0486) (0.0538) (0.213) (0.215) (0.216) (0.214) (0.0336)
Quarter dummies YES YES YES YES YES YES YES
Country dummies YES YES YES YES YES YES NO
Quarter-discipline dummies YES YES YES YES YES YES YES
(continued on next page)
F. Mueller-Langer, et al. Research Policy 49 (2020) 103886
16
Table 9 (continued)
(1) (2) (3) (4) (5) (6) (7)
VARIABLES OLS Base + Article info +Rank +Population +Dist._1 +Dist._2 +Inst.-Disc. FE
Institution-discipline FE NO NO NO NO NO NO YES
Observations 203,100 203,100 203,100 203,100 203,100 203,100 203,100
R-squared, overall 0.0917 0.0736 0.1044 0.1051 0.1060 0.1054 0.0571
Number of Inst_Discipline 4,062 4,062 4,062 4,062 4,062 4,062 4,062
Number of Inst 2,031 2,031 2,031 2,031 2,031 2,031 2,031
We use a balanced panel and take into account journal articles by single local authors only. Results on the impact of OARE membership (treated) on publication
output of research institutions in five developing countries (Bolivia, Ecuador, Kenya, Nigeria, Peru) from OLS DDD estimation. We use the xtreg command in STATA.
OLS regression coefficients reported. The institution-discipline-quarter triplets constitute the unit of observation. Period under study: 1st quarter 2000 to 2nd quarter
2012. Reference country is Nigeria. Reference quarter is 36. Reference rank is rank≤5000. Reference population is pop≤100. Random institution-discipline effects
are included in columns (1) to (6). Robust standard errors clustered at the institutional level reported in parentheses. *p< 0.1, **p< 0.05, ***p< 0.01.
Table 10
Effect of OARE on Publication Output (excluding Outliers).
(1) (2) (3) (4) (5) (6) (7)
Model: Base + Article info +Rank +Population +Dist._1 +Dist._2 +Inst.-Disc. FE
Dependent variable: w w w w w w w
OARE treated (DDD) 1.525*** 1.529*** 1.521*** 1.521*** 1.521*** 1.521*** 1.481***
(0.0895) (0.0892) (0.0883) (0.0883) (0.0883) (0.0883) (0.0853)
# pages 0.00211** 0.00165 0.00160 0.00161 0.00161 0.00263**
(0.00102) (0.00103) (0.00104) (0.00104) (0.00104) (0.00117)
# co-authors USA 0.00565 0.00535 0.00535 0.00537 0.00534 0.00318
(0.00490) (0.00466) (0.00465) (0.00466) (0.00466) (0.00463)
# co-authors EUR 0.0163*** 0.0160*** 0.0159*** 0.0160*** 0.0159*** 0.0173***
(0.00396) (0.00373) (0.00373) (0.00373) (0.00373) (0.00429)
Rank2: 5,000 < rank≤10,000 −0.769* −0.767* −0.771* −0.763*
(0.394) (0.394) (0.395) (0.395)
Rank3: 10,000 < rank≤15,000 −0.894** −0.876** −0.884** −0.876**
(0.348) (0.345) (0.345) (0.345)
Rank4: 15,000 < rank≤25,000 −0.0633 −0.0588 −0.0501 −0.0514
(0.513) (0.509) (0.511) (0.510)
Rank5: rank > 25,000 −1.029*** −1.042*** −1.040*** −1.038***
(0.334) (0.336) (0.337) (0.337)
Pop1: 100 < pop≤500, in 1,000 −0.0527 −0.0197 −0.0350
(0.0466) (0.0489) (0.0511)
Pop2: 500 < pop≤1,000, in 1,000 0.0565 0.0925 0.0700
(0.0631) (0.0691) (0.0681)
Pop3: 1,000 < pop≤5,000, in 1,000 0.0577 0.0350 0.0393
(0.0598) (0.0554) (0.0602)
Pop4: pop > 5,000, in 1,000 0.00428 −0.0874 −0.0334
(0.0341) (0.0596) (0.0384)
Distance from largest −0.0157**
domestic city, in 100 km (0.00794)
Distance from closest dom. city with > 1 −0.00838
mill. inhab., in 100 km (0.00593)
Constant 0.308*** 0.278*** 1.214*** 1.205*** 1.284*** 1.214*** 0.00205
(0.0577) (0.0582) (0.337) (0.338) (0.342) (0.338) (0.00929)
Quarter dummies YES YES YES YES YES YES YES
Country dummies YES YES YES YES YES YES NO
Quarter-discipline dummies YES YES YES YES YES YES YES
Institution-discipline FE NO NO NO NO NO NO YES
Observations 244,291 244,291 244,291 244,291 244,291 244,291 244,291
R-squared, overall 0.0662 0.0641 0.0738 0.0734 0.0734 0.0736 0.0869
Number of Inst_Discipline 4,980 4,980 4,980 4,980 4,980 4,980 4,980
Number of Inst 2,490 2,490 2,490 2,490 2,490 2,490 2,490
We use a balanced panel and take into account journal articles by both single and multiple local authors. Results on the impact of OARE membership (treated) on
publication output of research institutions in five developing countries (Bolivia, Ecuador, Kenya, Nigeria, Peru) from OLS DDD. Following Williams (2016), we delete
observations that have at least one of the following characteristics: (a) value of standardized residuals > 3, (b) leverage > 2k/nwhere kis the number of independent
variables in the regression and nis the number of observations, (c) Cook's Distance measure value > 4/n. In total, 4,709 observations are deleted. 1,355 observations
are deleted under (a), 13 under (b) and 3,341 under (c). We use the xtreg command in STATA. OLS regression coefficients reported. The institution-discipline-quarter
triplets constitute the unit of observation. Period under study: 1st quarter 2000 to 2nd quarter 2012. Reference country is Nigeria. Reference quarter is 36. Reference
rank is rank≤5000. Reference population is pop≤100. Random institution-discipline effects are included in columns (1) to (6). Robust standard errors clustered at
the institutional level. Note that serial correlation is not an issue in our balanced panel because the large number of periods with 0 publications breaks any time
correlation for any given institution. *p< 0.1, **p< 0.05, ***p< 0.01.
F. Mueller-Langer, et al. Research Policy 49 (2020) 103886
17
Appendix 1: Summary Statistics (Single Local Authors)
mean sd min max N
Dependent variables
# publications 0.112 0.848 0 55 203,100
# references 19.80 21.77 0 293 203,100
# OARE references 4.243 7.383 0 135 203,100
Countries
Kenya 0.256 0.436 0 1 203,100
Nigeria 0.240 0.427 0 1 203,100
Bolivia 0.134 0.341 0 1 203,100
Ecuador 0.137 0.344 0 1 203,100
Peru 0.233 0.423 0 1 203,100
Main variable of interest
OARE treated (DDD) 0.008 0.0860 0 1 203,100
Article characteristics
# co-authors USA 0.465 1.365 0 37.25 203,100
# co-authors EUR 0.569 1.727 0 54.33 203,100
# pages 6.081 6.547 0 73 203,100
Institutional characteristics
Rank1: rank≤5,000 0.0231 0.150 0 1 203,100
Rank2: 5,000 < rank≤10,000 0.0162 0.126 0 1 203,100
Rank3: 10,000 < rank≤15,000 0.0167 0.128 0 1 203,100
Rank4: 15,000 < rank≤25,000 0.0246 0.155 0 1 203,100
Table 11
Effect of OARE on Publication Output (Unbalanced Panel).
(1) (2) (3) (4) (5) (6) (7)
Model: Base + Article info +Rank +Pop. +Dist._1 +Dist._2 +Inst.-Disc. FE
Dependent variable: w w w w w w w
OARE treated (DDD) 4.951*** 4.916*** 3.723*** 3.663*** 3.654*** 3.661*** 2.944***
(1.152) (1.151) (0.890) (0.880) (0.875) (0.880) (0.729)
# pages −0.00344 −0.00225 −0.00367 −0.00306 −0.00321 −0.00159
(0.00719) (0.00710) (0.00651) (0.00662) (0.00672) (0.00352)
# co-authors USA 0.0966** 0.0938*** 0.0942*** 0.101*** 0.0942*** −0.00524
(0.0457) (0.0332) (0.0295) (0.0302) (0.0294) (0.0136)
# co-authors EUR 0.0356 0.0229 0.0160 0.0193 0.0177 0.0308
(0.0479) (0.0487) (0.0486) (0.0486) (0.0483) (0.0310)
Rank2: 5,000 < rank≤10,000 −1.242 −1.309 −1.208 −1.259
(1.327) (1.282) (1.299) (1.284)
Rank3: 10,000 < rank≤15,000 −2.143** −1.891* −2.011** −1.891*
(0.964) (0.980) (0.927) (0.971)
Rank4: 15,000 < rank≤25,000 0.531 0.588 0.561 0.605
(1.519) (1.427) (1.394) (1.421)
Rank5: rank > 25,000 −1.757** −1.984*** −2.010*** −1.978***
(0.724) (0.719) (0.716) (0.719)
Pop1: 100 < pop≤500, in 1,000 −1.238** −0.890* −1.144**
(0.527) (0.514) (0.576)
Pop2: 500 < pop≤1,000, in 1,000 0.218 0.469 0.275
(0.702) (0.712) (0.724)
Pop3: 1,000 < pop≤5,000, in 1,000 0.190 −0.0835 0.0596
(0.648) (0.595) (0.677)
Pop4: pop > 5,000, in 1,000 −0.456 −1.333* −0.717
(0.470) (0.691) (0.479)
Distance from largest domestic −0.161**
city, in 100 km (0.0771)
Distance from closest domestic city −0.0590
with > 1 million inhab., in 100 km (0.0792)
Constant 2.256*** 2.243*** 3.586*** 3.847*** 4.622*** 3.917*** 1.297***
(0.428) (0.413) (0.808) (0.831) (0.891) (0.839) (0.274)
Quarter dummies YES YES YES YES YES YES YES
Country dummies YES YES YES YES YES YES NO
Quarter-discipline dummies YES YES YES YES YES YES YES
Institution-discipline FE NO NO NO NO NO NO YES
Observations 16,131 16,131 16,131 16,131 16,131 16,131 16,131
R-squared, overall 0.1218 0.1251 0.1565 0.1657 0.1724 0.1662 0.0687
Number of Inst_Discipline 3,229 3,229 3,229 3,229 3,229 3,229 3,229
Number of Inst 2,490 2,490 2,490 2,490 2,490 2,490 2,490
We use an unbalanced panel and take into account journal articles by both single and multiple local authors. Results on the impact of OARE membership (treated) on
publication output of research institutions in five developing countries (Bolivia, Ecuador, Kenya, Nigeria, Peru) from OLS DDD. We use the xtreg command in STATA.
OLS regression coefficients reported. The institution-discipline-quarter triplets constitute the unit of observation. Period under study: 1st quarter 2000 to 2nd quarter
2012. Reference country is Nigeria. Reference quarter is 36. Reference rank is rank≤5000. Reference population is pop≤100. Random institution-discipline effects
are included in columns (1) to (6). Robust standard errors clustered at the institutional level. *p< 0.1, **p< 0.05, ***p< 0.01.
F. Mueller-Langer, et al. Research Policy 49 (2020) 103886
18
Rank5: rank > 25,000 0.919 0.272 0 1 203,100
Rank, in 1,000 6.709 5.044 0.749 21.24 203,100
City characteristics
Distance from largest domestic city, in 100 km 3.204 3.617 0 20.64 203,100
Distance from closest dom. city with > 1 million inhab., in 100 km 1.989 3.263 0 20.64 203,100
Pop0: pop≤100, in 1,000 0.200 0.400 0 1 203,100
Pop1: 100 < pop≤500, in 1,000 0.106 0.307 0 1 203,100
Pop2: 500 < pop≤1,000, in 1,000 0.161 0.368 0 1 203,100
Pop3: 1,000 < pop≤5,000, in 1,000 0.349 0.477 0 1 203,100
Pop4: pop > 5,000, in 1,000 0.184 0.387 0 1 203,100
We use a balanced panel and take into account journal articles by single local authors only. Data is aggregated at the institution-discipline-quarter level that
constitutes our unit of observation.
Appendix 2: Summary statistics by country Group
A.Summary statistics for Group A countries
mean sd min max N
Dependent variable
# publications 0.175 1.420 0 93.82 159,900
Countries
Kenya 0.398 0.490 0 1 159,900
Nigeria 0.401 0.490 0 1 159,900
Bolivia 0.201 0.401 0 1 159,900
Ecuador 0 0 0 0 159,900
Peru 0 0 0 0 159,900
Main variable of interest
OARE treated (DDD) 0.008 0.0864 0 1 159,900
Article characteristics
# co-authors USA 0.380 1.080 0 36 159,900
# co-authors EUR 0.499 1.633 0 57 159,900
# pages 6.064 6.047 0 66 159,900
Institutional characteristics
Rank1: rank≤5,000 0.0119 0.108 0 1 159,900
Rank2: 5,000 < rank≤10,000 0.00813 0.0898 0 1 159,900
Rank3: 10,000 < rank≤15,000 0.0188 0.136 0 1 159,900
Rank4: 15,000 < rank≤25,000 0.0369 0.189 0 1 159,900
Rank5: rank > 25,000 0.924 0.264 0 1 159,900
Rank, in 1,000 8.883 5.197 0.907 21.79 159,900
City characteristics
Distance from largest domestic city, in 100 km 3.523 3.568 0 20.64 159,900
Distance from closest dom. city with > 1 million inhab., in 100 km 2.031 2.915 0 20.64 159,900
Pop0: pop≤100, in 1,000 0.257 0.437 0 1 159,900
Pop1: 100 < pop≤500, in 1,000 0.0851 0.279 0 1 159,900
Pop2: 500 < pop≤1,000, in 1,000 0.226 0.418 0 1 159,900
Pop3: 1,000 < pop≤5,000, in 1,000 0.381 0.486 0 1 159,900
Pop4: pop > 5,000, in 1,000 0.0507 0.219 0 1 159,900
Data is aggregated at the institution-discipline-quarter level. The institution-discipline-quarter triplets constitute the unit of observation. We take into account journal
articles by both single and multiple local authors in the Group A countries under study. Registered research institutions receive free OARE membership in Group A
countries (GNI per capita below $1,600).
B. Summary statistics for Group B countries
mean sd min max N
Dependent variable
# publications 0.0768 0.598 0 29.53 89,100
Countries
Kenya 0 0 0 0 89,100
Nigeria 0 0 0 0 89,100
Bolivia 0 0 0 0 89,100
Ecuador 0.364 0.481 0 1 89,100
Peru 0.636 0.481 0 1 89,100
Main variable of interest
OARE treated (DDD) 0.006 0.0775 0 1 89,100
Article characteristics
# co-authors USA 0.637 1.673 0 37.25 89,100
# co-authors EUR 0.621 1.681 0 39 89,100
# pages 6.457 6.738 0 120 89,100
Institutional characteristics
Rank1: rank≤5,000 0.0348 0.183 0 1 89,100
Rank2: 5,000 < rank≤10,000 0.0303 0.171 0 1 89,100
Rank3: 10,000 < rank≤15,000 0.0168 0.129 0 1 89,100
F. Mueller-Langer, et al. Research Policy 49 (2020) 103886
19
Rank4: 15,000 < rank≤25,000 0.00449 0.0669 0 1 89,100
Rank5: rank > 25,000 0.914 0.281 0 1 89,100
Rank, in 1,000 4.137 1.714 0.749 21.39 89,100
City characteristics
Distance from largest domestic city, in 100 km 2.771 3.643 0 15.73 89,100
Distance from closest dom. city with > 1 million inhab., in 100 km 1.920 3.693 0 15.73 89,100
Pop0: pop≤100, in 1,000 0.153 0.360 0 1 89,100
Pop1: 100 < pop≤500, in 1,000 0.145 0.352 0 1 89,100
Pop2: 500 < pop≤1,000, in 1,000 0.0393 0.194 0 1 89,100
Pop3: 1,000 < pop≤5,000, in 1,000 0.251 0.434 0 1 89,100
Pop4: pop > 5,000, in 1,000 0.412 0.492 0 1 89,100
Data is aggregated at the institution-discipline-quarter level. The institution-discipline-quarter triplets constitute the unit of observation. We take into account journal
articles by both single and multiple local authors in the Group B countries under study. Registered research institutions receive reduced-fee OARE membership
($1,000 per year) in Group B countries (GNI per capita below $5,000).
Appendix 3: Histogram of the number of publications
Histogram of number of publications, w, at the institution-discipline-quarter level. wis the dependent variable in the regressions.
Appendix 4: Bayesian methodology
For robustness, we also estimate the OARE effect using Bayesian estimation techniques based on a data augmentation MCMC algorithm. There are
two equations. The first equation determines self-selection in the OARE initiative using a latent variable framework. The second equation is identical
to Eq. (1). We assume that the unobserved variables of both equations follow a bivariate normal distribution with correlation coefficient ρ. The
MCMC algorithm simulates the latent variable of the first equation to generate the endogenous binary treatment effect. The Bayesian approach
explicitly deals with the correlation between the unobserved variables of the two equations. If there are any unobserved variables that determine
whether an institution self-selects into the OARE program, the Bayesian method accounts for its potential endogeneity on the estimation of the
treatment effect. This comes at the cost of increasing computing time since the self-selection and productivity equations are estimated at the same
time.
Eq. (A1) determines the outcome of the endogenous binary variable:
=
>
yif w
if w
1, 0
0, 0
i
i
i
1,
1,
1,
(A1)
where w
1,i
= x
1,i
′ β
1
+ε
1,i
, β
1
is of dimension k
1
and x
1,i
is a set of k
1
control variables.
Eq. (A2) explains the observed variable w
2,i
as a function of individual characteristics and the endogenous binary variable z
1,i
= 1 if w
1,i
> 0 and
z
1,i
= 0 if w
1,i
≤ 0,
(A2)
where δ
1
is the structural parameter associated with the binary endogenous variable z
1
, z
2,i
is a set of k
2
explanatory variables not necessary identical
to x
1,i
and δ
2
is a vector of parameters of dimension k
2
, x
2,i
= (z
1,i
, z
2,i
′)′ and β
2
= (δ
1
, δ
2
′)′. We assume that
=( , )
ii i1, 2,
is normally distributed
with mean (0, 0)′ and covariance Σfor i= 1, …, n:
=
1
2
. Parameter ρrepresents the correlation between the unobservable variables.
Parameter σ
2
is the variance of ɛ
2, i
. Since the probit Eq. (A1) is not identified, we chose to normalize the variance of the endogenous binary variable
to 1. This is a standard restriction in probit models.
Let β = (β
1
′, β
2
′)′, w
1
= (w
1,1
, …, w
1,n
)′, w
2
= (w
2,1
, …, w
2,n2
)′ and define w = (w
1
′, w
2
′)′. We define ε
1
,ε
2
, and εin a similar fashion.
The covariance of the unobservable variables is simply
= =E In
where I
n
denotes the identity matrix of dimension n×n. Thus Ω
−1
is readily obtained. We similarly define
F. Mueller-Langer, et al. Research Policy 49 (2020) 103886
20
= × +Xx
xn k k
0
02 ( )
11
21 2
The (partially) latent model can be written in matrix format:
= +w X
(A3)
Hence conditional on w and Ω, the estimates of β are simply obtained by a generalized least-squares (GLS) regression of (A3).
38
Moreover, the
matrices X′Ω
−1
X and X′Ω
−1
w required for the GLS estimates of the parameters of the model are easily computed. We use a uniform prior for β, ρand
a non-informative prior for σ: p(β, ρ, σ)1/σ.
39
The Metropolis-Gibbs sampling algorithm proceeds in 4 steps drawing from conditional distributions
sequentially. The full procedure is described in Bounie et al. (2016). Parameters of the model are identified by the non-linearity of the probability of
observing the endogenous equal to me, just as in the sample selection model. There is however no sample selection since we have observations for
the two equations for all institutions in our sample. Moreover, the model is triangular and thus satisfies the principal assumption of Heckman (1978).
Self-selection is accounted by the correlation between the unobservable self-selection equation and the productivity equation.
Appendix 5: OARE effect by country group
(1) (2) (3) (4) (5) (6)
Country Group: Group A Group B
Model: distance_1 distance_2 Inst.-Disc. FE distance_1 distance_2 Inst.-Disc. FE
Dependent variable: w w w w w w
OARE treated (DDD) 2.472*** 2.472*** 2.446*** 1.239*** 1.239*** 1.213***
(0.542) (0.542) (0.536) (0.409) (0.409) (0.402)
# pages 0.00628* 0.00628* 0.00873* −0.000539 −0.000539 0.000764
(0.00358) (0.00358) (0.00474) (0.00471) (0.00471) (0.00614)
# co-authors USA 0.00842 0.00840 0.0102 0.00182 0.00181 0.000508
(0.0151) (0.0151) (0.0172) (0.0116) (0.0116) (0.0129)
# co-authors EUR −0.00676 −0.00683 −0.00626 0.122** 0.122** 0.141**
(0.00906) (0.00905) (0.0101) (0.0531) (0.0531) (0.0567)
Rank2: −0.679 −0.666 −0.844*** −0.845***
5,000 < rank≤10,000 (0.959) (0.958) (0.283) (0.283)
Rank3: −1.214* −1.195 −0.790*** −0.791***
10,000 < rank≤15,000 (0.726) (0.729) (0.262) (0.262)
Rank 4: −0.298 −0.293 −0.892*** −0.892***
15,000 < rank≤25,000 (0.845) (0.848) (0.286) (0.286)
Rank5: −1.359* −1.346* −0.815*** −0.815***
rank > 25,000 (0.715) (0.717) (0.259) (0.260)
Pop1: 0.000493 −0.0373 −0.0643* −0.0656*
100 < pop≤500, in 1,000 (0.0767) (0.0789) (0.0345) (0.0348)
Pop2: 0.150 0.0931 −0.0300 −0.0302
500 < pop≤1,000, in 1,000 (0.0944) (0.0938) (0.0341) (0.0342)
Pop3: 0.0189 0.0602 0.0462 0.0503
1,000 < pop≤5,000, in 1,000 (0.0746) (0.0883) (0.0461) (0.0466)
Pop4: −0.265*** −0.128** 0.0378 0.0395
pop > 5,000, in 1,000 (0.103) (0.0653) (0.0231) (0.0242)
Distance from largest domestic city, in 100 km −0.0279** 0.00189
(0.0127) (0.00232)
Distance from closest dom. city with −0.00610 0.00220
> 1 mill. inh., in 100 km (0.0127) (0.00228)
Constant 1.675** 1.524** 0.161*** 0.793*** 0.792*** −0.0646
(0.708) (0.704) (0.0349) (0.259) (0.258) (0.0431)
Quarter dummies YES YES YES YES YES YES
Country dummies YES YES NO YES YES NO
Quarter-discipline dummies YES YES YES YES YES YES
Institution-discipline FE NO NO YES NO NO YES
Observations 159,900 159,900 159,900 89,100 89,100 89,100
R-squared, overall 0.1178 0.1162 0.0969 0.1333 0.1333 0.0584
Number of Inst_Discipline 3,198 3,198 3,198 1,782 1,782 1,782
Number of Inst 1,599 1,599 1,599 891 891 891
We use a balanced panel and take into account journal articles from both single and multiple local authors. Results on the impact of OARE membership (OARE
treated) on publication output of research institutions by country group (Group A: Bolivia, Nigeria, Kenya; Group B: Ecuador, Peru) from OLS DDD estimation. We use
Stata's xtreg command. OLS regression coefficients reported. The institution-discipline-quarter triplets constitute the unit of observation. Period under study: 1st
quarter 2000 to 2nd quarter 2012. Reference quarter is 36. Reference rank is rank≤5000. Reference population is pop≤100. Random institution-discipline effects are
included in columns (1), (2), (4), and (5). Robust standard errors clustered at the institutional level. Note that serial correlation is not an issue in our balanced panel
because the large number of periods with 0 publications breaks any time correlation for any given institution. *p< 0.1, **p< 0.05, ***p< 0.01.
38
Since each stage generally includes different sets of explanatory variables, we cannot estimate the seemingly unrelated regressions model with ordinary least-
squares regression applied to each latent equation separately.
39
The choice of the prior distribution does not matter much when there is a large number of observations. Moreover, using the uniform prior distribution provides
a direct means of comparison with the maximum likelihood procedures.
F. Mueller-Langer, et al. Research Policy 49 (2020) 103886
21
Appendix 6: OARE effect by country
(1) (2) (3) (4) (5)
Model: +Inst.-Disc. FE +Inst.-Disc. FE +Inst.-Disc. FE +Inst.-Disc. FE +Inst.-Disc. FE
Country: Kenya Nigeria Bolivia Ecuador Peru
Dependent variable: w w w w w
OARE treated (DDD) 2.017*** 2.773*** 0.811** 0.840* 1.364**
(0.580) (0.728) (0.380) (0.435) (0.550)
# pages 0.00952 0.0122 0.00356 −0.0116 0.00877
(0.00679) (0.00969) (0.00455) (0.00751) (0.00592)
# co-authors USA 0.0209 −0.0395 0.00804 −0.0291 −0.00821
(0.0221) (0.0498) (0.0172) (0.0498) (0.0173)
# co-authors EUR −0.00597 −0.0195 0.00511 0.276*** 0.0780**
(0.0195) (0.0442) (0.00922) (0.0938) (0.0394)
Constant 0.0472 0.0636 0.0345 −0.0703 −0.0374
(0.0546) (0.0679) (0.0305) (0.0479) (0.0601)
Quarter dummies YES YES YES YES YES
Country dummies NO NO NO NO NO
Quarter-discipline dummies YES YES YES YES YES
Institution-discipline FE YES YES YES YES YES
Observations 63,700 64,100 32,100 32,400 56,700
R-squared, within 0.050 0.086 0.035 0.162 0.083
Number of Inst_Discipline 1,274 1,282 642 648 1,134
Number of Inst 637 641 321 324 567
We use a balanced panel and take into account journal articles from both single and multiple local authors. Results on the impact of OARE membership (treated) on
publication output of research institutions by country from OLS DDD estimation. We use Stata's xtreg command. OLS regression coefficients reported. The institution-
discipline-quarter triplets constitute the unit of observation. Period under study: 1st quarter 2000 to 2nd quarter 2012. Reference quarter is 36. Reference rank is
rank≤5000. Reference population is pop≤100. Robust standard errors clustered at the institutional level. Note that serial correlation is not an issue in our balanced
panel because the large number of periods with 0 publications breaks any time correlation for any given institution. *p< 0.1, **p< 0.05, ***p< 0.01.
References
Acs, Z.J., Audretsch, D.B., Feldman, M.P., 1994. R&D spillovers and recipient firm size.
Rev. Econ. Stat. 76 (2), 336–340.
Annan, K., 2004. Science for all nations. Science 303, 925.
Audretsch, D.B., Feldman, M.P., 1996. R&D spillovers and the geography of innovation
and production. Am. Econ. Rev. 86 (3), 630–640.
Bergstrom, T., Bergstrom, C.T., 2004. The costs and benefits of library site licenses to
academic journals. Proc. Natl. Acad. Sci. 101 (3), 897–902.
Bounie, D., Abel, F., Waelbroeck, P., 2016. Debit card and demand for cash. J. Bank.
Financ 73, 55–66.
Dasgupta, P., David, P.A., 1994. Toward a new economics of science. Res. Policy 23 (5),
487–521.
Davis, P.M., 2011. Do discounted journal access programs help researchers in Sub-
Saharan Africa? A Bibliometric Analysis. Learned Publishing 24 (4), 287–298.
Dosi, G., 1988. Sources, procedures, and microeconomic effects of innovation. J. Econ.
Lit. 26 (3), 1120–1171.
Eger, T., Scheufen, M., Meierrieks, D., 2015. The determinants of open access publishing:
Survey evidence from Germany. Eur. J. Law Econ. 39 (3), 475–503.
European Commission, 2012. Towards better access to scientific information: Boosting
the benefits of public investments in research. Brussels: European Commission.
Evans, J.A., Reimer, J., 2009a. Open access and global participation in science. Science
323, 1025.
Evans, J.A., Reimer, J., 2009b. Open access and global participation in science.
Supporting Online Material. Science. pp. 323.
Frandsen, T.F., 2009. Attracted to open access journals: A bibliometric author analysis in
the field of biology. J. Document. 65 (1), 58–82.
Furman, J.L., Stern, S., 2011. Climbing atop the shoulders of giants: The impact of in-
stitutions on cumulative research. Am. Econ. Rev. 101 (5), 1933–1963.
Gaulé, N., Maystre, N., 2011. Getting cited: Does open access help? Res. Policy 40 (10),
1332–1338.
Gaulé, P., 2009. Access to scientific literature in India. J. Am. Soc. Inform. Sci. Tech. 60
(12), 2548–2553.
Geroski, P.A., 2000. Models of technology diffusion. Res. Policy 29 (4–5), 603–625.
Griliches, Z., 1957. ‘Hybrid corn’: An exploration in the economics of technological
change. Econometrica 25 (4), 501–522.
Griliches, Z., 1992. The search for R&D spillovers. Scand. J. Econ. 94, 29–47.
Hall, B., 2004. Innovation and diffusion, NBER, Cambridge, MA NBER Working Paper No.
10212.
Hall, B., Khan, B., 2003. Adoption of new technology, NBER. Cambridge, MA NBER
Working Paper(9730) No.
Hanauske, M., Bernius, S., Dugall, B., 2007. Quantum game theory and open access
publishing. Physica A. 382, 650–664.
Heckman, J.J., 1978. Dummy endogenous variables in a simultaneous equation system.
Econometrica 46 (4), 931–959.
Jaffe, A.B., 1989. Real effects of academic research. Am. Econ. Rev. 79 (5), 957–970.
Jeon, D.S., Rochet, J.C., 2010. The pricing of academic journal: A two-sided market
perspective. Am. Econ. J. Microecon. 2, 222–255.
Mansfield, E., 1961. Technical change and the rate of imitation. Econometrica 29 (4),
741–766.
Mansfield, E., 1963. Intrafirm rates of diffusion of an innovation. Rev. Econ. Stat. 45 (4),
348–359.
Merton, R.K., 1973. The sociology of science: theoretical and empirical investigations,
Storer, N.W. (editor), University of Chicago Press: Chicago, IL.
McCabe, M.J., Snyder, C.M., 2005. Open access and academic journal quality. Am. Econ.
Rev. Paper Proc. 95 (2), 453–458.
McCabe, M.J., Snyder, C.M., 2014. Identifying the effect of open access on citations using
a panel of science journals. Econ. Inq. 52 (4), 1284–1300.
McCabe, M.J., Snyder, C.M., 2015. Does online availability increase citations? Theory and
evidence from a panel of economics and business journals. Rev. Econ. Stat. 97 (1),
144–165.
McCabe, M.J., Snyder, C.M., 2018. Open access as a crude solution to a hold-up problem
in the two-sided market for academic journals. J. Indust. Econ. 66 (2), 239–517.
Mueller-Langer, F., Scheufen, M., 2013. Academic publishing and open access. In: Towse, R.,
Handke, C. (Eds.), Handbook of the Digital Creative Economy. Edward Elgar, pp. 365–377.
Mueller-Langer, F., Watt, R., 2018. How many more cites is a $3,000 open access fee buying
you? Empirical evidence from a natural experiment. Econ. Inq. 56 (2), 931–954.
Murray, F.E., Aghion, P., Dewatripont, M., Kolev, J., Stern, S., 2016. Of mice and aca-
demics: Examining the effect of openness on innovation. Am. Econ. J. Econ. Policy 8
(1), 212–252.
Ramello, G.B., 2010. Copyright and endogenous market structure: A glimpse from the
journal-publishing market. Rev. Econ. Res. Copyright Issues 7 (1), 7–29.
Romer, P.M., 1986. Increasing returns and long-run growth. J. Polit. Econ. 94 (5),
1002–1037.
Romer, P.M., 1990. Endogenous technological change. J. Polit. Econ. 98 (5), 71–102.
Scheufen, M., 2015. Copyright versus open access: on the organisation and international po-
litical economy of access to scientific knowledge, Springer International Publishing.
Shavell, S., 2010. Should copyright for academic works be abolished? J. Legal Anal. 2 (1),
301–358.
Solow, R.M., 1956. A contribution to the theory of economic growth. Q. J. Econ. 70 (1),
65–94.
Sorenson, O., Fleming, L., 2004. Science and the diffusion of knowledge. Res. Policy 33
(10), 1615–1634.
Stephan, P.E., 1996. The economics of science. J. Econ. Lit. 34 (3), 1199–1235.
Suber, P., Arunachalam, S., 2005. Open access to science in the developing world, http://dash.
harvard.edu/bitstream/handle/1/4725025/suber_worldinfocity.htm?sequence=1.
Suber, P., 2012. Open access. Cambridge: MIT Press.
Swan, T.W., 1956. Economic growth and capital accumulation. Econ. Rec. 32 (2), 334–361.
Williams, R., 2016. Outliers, University of Notre Dame, available at: https://www3.nd.
edu/~rwilliam/.
F. Mueller-Langer, et al. Research Policy 49 (2020) 103886
22
... In the last two decades, the Internet has become the non-existent into the largest and most accessible database of information ever created (Mueller-Langer et al., 2020). The Internet has changed the way people communicate with others, purchase items/ products, socialize, gain knowledge, and learn. ...
Article
Online education has gained immense popularity among the working people and students pursuing higher education. Various renowned universities all over the world are offering online degrees and diplomas to all people through digital technologies. This enhances the concept of online classes due to the complete shutdown of educational institutions for an indefinite COVID-19 pandemic situation. Though the online classes are an immediate and emergency paradigm shifting in teaching and learning, it has certain drawbacks that concern the student to a larger extent. To examine the effects of online classes in terms of quality, comfort, and compatibility, this study analyzes the students' perception of various arts, science, and engineering colleges. Drawing on data from various students and its statistical test has inferred various challenges faced by the students with respect to their discipline and living locality. The result of statistical analysis recommends more improvisations and special considerations to the educational institutions to make this mode a viable solution.
... n.d.). In addition to such access being valuable for supporting teaching at institutions, research has shown that access provided through such programs to developing countries has an impact in facilitating published research conducted in the region (Mueller-Langer et al., 2020). However, barriers for interacting with the global scholarly literature are not only related to costs and subscriptions, but poor access to the internet, frequent blackouts, poor information technology infrastructure, and lack of skills (Raju et al., 2020). ...
Article
Full-text available
Open access (OA) to research publications is of global relevance, both in terms of provision and consumption of scholarly content. However, much of the research, practice, and models surrounding OA have been centered around the Global North. In this study we investigate how and to what degree higher education institutions (HEIs) in Angola interact with the concept of OA to journal publications through their policies and practices, a country where the end of the civil war in 2002 marked a new start for growth in teaching and research. This study is based on an online survey conducted in 2020 among research management units of Angolan HEIs. 23 valid institutional responses were received of 44 invitations sent (52% response rate). The results suggest that Angolan HEIs have moderate awareness of OA but practical incorporation into academic processes has remained slow, however, this can be seen to be connected to the overall slow progress in ramping up research intensity in the country. Seven of the responding institutions reported to be involved in publishing scholarly journals, all of them OA. Overall Angolan HEIs have few institutional repositories, and have so far placed little value on OA in the context of academic career advancement.
... The OA regime can thus promote scientific participation and thereby foster the global evolution of science as a "trial and error" process. Free or cheaper access to literature for researchers in the developing world tends to increase both their output (number of publications) and input (number of references) (Mueller-Langer et al., 2020). Second, an OA regime grants access to the latest results in science for groups who were previously excluded because they are not "club" members of a university library. ...
... Although the question of the impact of digital transformations on the publishing market is largely invested in the literature, little work analyzes the question from an economic point of view (e.g. market balance). Much of the current research on the publishing market reports on the various existing models and their development (Mazzarol and Soutar, 2001;Bernius et al., 2009;Laakso et al., 2011;Scheufen, 2015;Potts et al., 2017;Magadán-Díaz and Rivas-García, 2018;Clark and Phillips, 2019), characteristics of the competition (Feather, 2006;Björk, 2017b;Ponte, Mierzejewska and Klein, 2017;Hagner, 2018), costs of scientific publishing in the OA era (McCabe, Snyder and Fagin, 2013;Van Noorden, 2013;Pinfield, Salter and Bath, 2016;Bruns, Rimmert and Taubert, 2020), self-archiving practices for research results (Alt, Militzer-Horstmann and Zimmermann, 2016a;Vuong, 2020), the difficulties to switch to an OA model and the negotiation power of publishers, or the future challenges of scientific publishing (Butler, 2013;Monastersky, 2013;Wilbanks, 2013;Lomazzi and Chartron, 2014;Björk, 2017b;Asai, 2020aAsai, , 2020b, including in emerging countries (Schöpfel, 2017;Cassi, Dosso and Mescheba, 2018;Iyandemye and Thomas, 2019;Vuong, 2019;Mueller-Langer, Scheufen and Waelbroeck, 2020;Raju et al., 2020Raju et al., , 2020. ...
Preprint
Full-text available
Stakeholders in research and scientific publishing are gradually joining the Open-Access (OA) movement, which is gaining momentum to become nowadays at the heart of scientific policies in high-income countries. The rise of OA generates profound changes in the chain of production and dissemination of knowledge. Free access to peer-reviewed research methods and results has contributed to the dynamics of science observed in recent years. The modes of publication and access have also evolved; the classic model, based on journal subscriptions is gradually giving way to new economic models that have appeared with the arrival of OA. The objective of this article is twofold. First, propose a model for the publishing market based on the literature as well as on changes in open science policies. Second, analyze publishing strategies of publishers and institutions. To do so, we relied on game theory in economics. Results show that in the short term, the publisher's equilibrium strategy is to adopt a hybrid-publishing model, while the institutions' equilibrium strategy is to publish in OA. This equilibrium is not stable and that in the medium/long term, the two players will converge on an OA publishing strategy. The analysis of the equilibrium in mixed-strategies confirms this result.
... OA transcends the initial motivator of public good to other objectives such as boosting research impact, increasing institutional visibility, and guaranteeing long-term costeffective access to research. Mueller-Langer et al. (2020) have demonstrated that providing OA to journals boosts research output by up to 29.6%. There are increasing indicators that HEIs in the UAE are striving to achieve all or some of these goals. ...
Article
Full-text available
Higher education institutions (HEIs) have an instrumental role in the move towards Open Access (OA) by shaping the national strategies, policies, and agendas. This study sets out to explore the role of HEIs in the United Arab Emirates (UAE) OA uptake and reflect on the ongoing international initiatives pushing for universal OA to research. The study is based on an online survey targeted at UAE higher education institutions research management units. In order to measure the institutional views, only one response was solicited from each institution. A total of 19 valid responses were received, making up 47% of HEIs included in the population of organisations. Our results suggest that there is low commitment to OA among UAE HEIs as attested by the low number of OA policies, scarce OA funding, limited proliferation of institutional repositories, perceived lack of urgency to migrate from current access models, and little consideration of OA for promotion purposes. The study is the first of its kind in the UAE, Arab and Middle Eastern countries, providing rare insight into a growing phenomenon that is global, yet most vocally discussed from a western perspective and context. The study contributes to the debate on the role of HEIs in the transition to OA and in shaping national and regional OA policies, as well as informing international initiatives about the current status of OA in the region.
... Since the turn of this century, there were massive changes in scholarly communication due to digitalisation and therefore, scientific material was easily accessible from the internet (Baffy et al., 2020). As a consequence, free access to research articles through the internet has increased the scientific output from developing countries (Mueller-Langer, Scheufen, & Waelbroeck, 2020). This has raised a serious concern about the quality of research articles published through online open access journal systems (Bohannon, 2013). ...
Article
Full-text available
The coronavirus 2019 (COVID-19) disease has affected millions of lives, forcing most of us to stay at home and work. However, there is an immediate need to conduct research on potential drugs against COVID-19. In this article, the extent to which major publishers have provided access for the public to read research articles relevant to potential drug candidates for the COVID-19 disease are presented. A systematic search of five electronic databases (Elsevier’s ScienceDirect, Taylor & Francis, SpringerLink, Wiley, and New England Journal of Medicine (NEJM)) was conducted on April 12-17, 2020. The total number of research articles containing terms ‘Ribavirin’, ‘Remdesivir’, ‘Hydroxychloroquine OR Chloroquine’, ‘Favipiravir’, ‘Lopinavir OR Ritonavir’, ‘Sarilumab’, and ‘Tocilizumab’, available for the public to read for free were determined. In this study, there was a lack of full access to research articles related to potential drugs of COVID-19 in commercial academic databases, except for ‘Remdesivir’ and ‘Favipiravir’ from NEJM.
Article
Full-text available
Does open access (OA) to journal articles foster citations to these articles? We compare the citation impact of gold and green OA in two disciplines: Biology and Economics & Management. The empirical analysis covers all articles of these disciplines included in the Web of Science “Journal Citation Reports” between 2000 and 2019. We show that, controlling for confounding variables pertaining to the journals and articles, gold OA increases citations across all articles. However, the individual disciplines feature starkly different effects: a 18.3% increase in Biology, compared to a decrease by 30.9% in Economics & Management. Also Green OA leads to an increase in citations to academic research. These results are confirmed by a number of robustness checks.
Article
Full-text available
This chapter examines the contributions that economists have made to the study of science and the types of contributions the profession is positioned to make in the future. Special emphasis is placed on the public nature of knowledge and characteristics of the reward structure that encourage the production and sharing of knowledge. The role that cognitive and noncognitive resources play in discovery is discussed as well as the costs of resources used in research. Different models for the funding of research are presented. The chapter also discusses scientific labor markets and the extreme difficulty encountered in forecasting the demand for and supply of scientists. The chapter closes with a discussion of the relationship of scientific research to economic growth and suggestions for future research.
Article
We analyze a model in which journals cannot commit to subscription fees when authors (who prefer low subscription fees because this boosts readership) make submission decisions. A hold‐up problem arises, manifested as excessive subscription fees. Open access is a crude attempt to avoid hold up by eliminating subscription fees. We assess the profitability and efficiency of traditional relative to open‐access journals in a monopoly model (with extensions to nonprofit, bundled, hybrid, and competing journals). We apply the theory to understand the evolving market for academic journals in the Internet age and policies currently being debated such as an open‐access mandate.
Article
We analyze the effect of open access (OA) status of journal articles on citations. Using cross-sectional and panel data from mathematics and economics, we perform negative binomial, Poisson, and generalized method of moments/instrumental variable methods regressions. We benefit from a natural experiment via hybrid OA pilot agreements. Citations to pre-prints allow us to identify the intrinsic quality of articles prior to journal publication. Overall, our analysi