Air travel and research collaboration: a quasi-experimental insight

Abstract

This article analyzes the impact of the availability of long-haul flights on international scientific collaboration. The background assumption is that a wider availability of long-distance flights enables greater mobility for scientists, which in turn increases the likelihood of long-distance research collaboration. The analytical framework of this article is based on a quasi-experimental approach. Specifically, a discontinuity in the global network of air routes due to regulatory requirements is used as a source of random selection—i.e., a natural experiment. Two complementary methods are used to model the effect: instrumental variable design and regression discontinuity. Combining the bibliometric data from Microsoft Academic Graph and flight data from the International Civil Aviation Organization, this paper provides evidence of a causal relation between air transport availability and research collaboration. The results show that direct long-haul flights positively impact the number of papers co-authored by scholars based in both the connected cities. The implications of these results for science policy are discussed in the context of the virtualization of scientific communication resulting from the COVID-19 pandemic and air transport’s negative climate impacts.

Full text

Vol.:(0123456789)
Scientometrics
https://doi.org/10.1007/s11192-025-05291-5
Air travel andresearch collaboration: aquasi‑experimental
insight
AdamPloszaj1
Received: 31 May 2024 / Accepted: 14 March 2025
© The Author(s) 2025
Abstract
This article analyzes the impact of the availability of long-haul flights on international sci-
entific collaboration. The background assumption is that a wider availability of long-dis-
tance flights enables greater mobility for scientists, which in turn increases the likelihood
of long-distance research collaboration. The analytical framework of this article is based
on a quasi-experimental approach. Specifically, a discontinuity in the global network of
air routes due to regulatory requirements is used as a source of random selection—i.e.,
a natural experiment. Two complementary methods are used to model the effect: instru-
mental variable design and regression discontinuity. Combining the bibliometric data from
Microsoft Academic Graph and flight data from the International Civil Aviation Organiza-
tion, this paper provides evidence of a causal relation between air transport availability and
research collaboration. The results show that direct long-haul flights positively impact the
number of papers co-authored by scholars based in both the connected cities. The implica-
tions of these results for science policy are discussed in the context of the virtualization of
scientific communication resulting from the COVID-19 pandemic and air transport’s nega-
tive climate impacts.
Keywords Research collaboration· Academic mobility· Air travel· Natural experiment·
Quasi experiment
Introduction
As a social practice, contemporary science can be characterized by widespread collabo-
ration and internationalization. Numerous studies have documented the development of
international research collaboration, academic mobility, and knowledge flows, as well as
their patterns, causes, and consequences (Olechnicka etal., 2019). The dominant view is
that scientific mobility is beneficial for individual scientists and their careers, as well as
for scientific institutions and general advancement (Hall etal., 2018; Jacob & Meek, 2013;
Leahey, 2016; Netz et al., 2020; Petersen, 2018). This topic is important not only from
* Adam Ploszaj
a.ploszaj@uw.edu.pl
1 Science Studies Lab & Centre forEuropean Regional andLocal Studies EUROREG, University
ofWarsaw, Warsaw, Poland

Scientometrics
a theoretical point of view but also from the perspective of scientific and, more broadly,
development policy. Many institutions and countries have specific policies supporting
scientific mobility based on the assumption that it can yield positive results, including
such iconic initiatives as the Fulbright in the United States (Arndt, 1996) and the Marie
Skłodowska-Curie Actions in the European Union (Souto-Otero, M etal., 2017).
At the same time—as in the case of many, if not all, socioeconomic issues—conduct-
ing a strict causal analysis of the effects of scientific mobility is highly complicated, or
even impossible. Indeed, it is difficult to imagine an experimental study in which fate alone
decides which scientist obtains a scholarship to go abroad and which stays in their own
country. Cost would make such an experiment impossible, not to mention ethical issues.
This leaves two possibilities. The first is observational research, which by definition pro-
hibits the confirmation or rejection of causal hypotheses. The second option is a quasi-
experimental approach, which does not involve an experiment in the strict sense. Instead,
appropriate techniques are applied to suitable data and specific assumptions are fulfilled,
allowing conclusions to be drawn about cause and effect from a given analysis (Campbell
& Stanley, 2011). The quasi-experimental approach is often based on the occurrence of
certain unique circumstances that introduce a random element and, consequently, divide
the examined sample into an experimental and a control group. In such a situation, we
speak of natural experiments. A natural experiment differs from a proper experiment: in
the latter, the random element is introduced as part of the research procedure, while in the
former the random factor is outside the researchers’ control (which also cleverly removes
all problematic legal and ethical issues). However, in practice, the use of natural experi-
ments is limited because finding an appropriate quasi-experimental situation is more dif-
ficult than one might expect (Dunning, 2012; Sekhon & Titiunik, 2012).
This article uses a highly specific quasi-experimental situation to analyze the impact of
long-distance mobility on research collaboration. In particular, the article analyzes how the
availability of long-distance (generally speaking: intercontinental) air connections trans-
lates into scientific collaboration. An important assumption here is that the physical pos-
sibility of travel materializes in the form of traveling scholars (scientific mobility). Accord-
ingly, scientists can meet and collaborate, leading to measurable effects in the form of
co-authored publications. The key element that translates into materialized collaboration
is thus not the availability of transport infrastructure but rather the face-to-face meetings it
enables.
Theoretically, this study is embedded in two key concepts explaining the spatial aspects
of research collaboration. The first is the importance of spatial distance for establishing,
maintaining, and conducting scientific collaboration. The basic rule in this case is that
the likelihood of collaboration declines as the distance between prospective collaborators
increases. This thesis finds support in a large body of published research at the micro (e.g.,
collaboration likelihood within buildings or campuses), meso (e.g., cities, regions, coun-
tries), and macro (e.g., global) scales (Allen, 2007; Fernández et al., 2016; Kabo etal.,
2014; Katz & Martin, 1997). The second concept embeds research collaboration in the
gravity model borrowed from physics, adding a mass variable to the distance variable.
Here, mass is typically interpreted as research capacities, measured, for instance, using the
volume of published papers, submitted patents, received grants, or research workforce. In
effect, the basic gravity model theoretical framework can be written as follows: The prob-
ability and intensity of research collaboration are negatively associated with geographical
distance, which separates the units in question, and positively affected by their accumulated
research potential. Again, numerous studies have used this approach (See e.g.: Andersson
& Persson, 1993; Hoekman etal., 2009; Picci, 2010; Plotnikova & Rake, 2014; Sebestyén

Scientometrics
& Varga, 2013). This article uses the gravity framework, while deepening and improving
our understanding of the phenomenon under study by employing a quasi-experimental
design and including variables describing actual long-distance transport accessibility as a
factor enabling scientific mobility, which in turn impacts research collaboration.
The analyses presented in this article seek to test the hypothesis that greater availability
of long-distance air connections positively influences long-distance research collaboration.
A key aspect of the hypothesis is the centrality of a variable that is not directly observed,
namely the mobility of scientists. The existence of air connections should not (based on
existing theories) directly translate into research collaboration. However, the availability
of air connections facilitates this mobility in the short term, such as for conference trips or
project meetings, as well as in the medium and long term, such as internships, study vis-
its, and international fellowships. The mobility of scientists enables face-to-face meetings,
which significantly increases the chance of establishing collaboration (Boudreau etal.,
2017; Chai & Freeman, 2019; Duede etal., 2024; Jin etal., 2024). As such, the study’s
conceptual framework can be presented as follows: (1) Better availability of long-distance
flights (observed variable) influences (2) greater mobility of scientists (unobserved vari-
able), which in turn increases the likelihood of (3) long-distance research collaboration
(observed variable). Scientific collaboration is operationalized as the co-authorship of sci-
entific publications. Of course, this is not the only form or indication of scientific collabo-
ration, but it is a commonly used measure due to both the availability and large volume of
data, which facilitates statistical analysis by providing large samples.
Prior work
This study fits directly into the field of transport accessibility and its effects on academic
mobility, research collaboration, and scientists’ productivity. It is worth noting that few
analyses of mobility and scientific collaboration consider transport accessibility, espe-
cially given the hundreds of publications devoted to the topic of collaboration in science.
Even when included, spatial separation is often simply measured as geographical distance
along the surface of the earth (or colloquially: “as the crow flies”) (Frenken etal., 2009a,
2009b). Furthermore, while a handful of studies have accounted for transport accessibil-
ity, they have typically been descriptive (i.e., non-causal) and focused on specific cases,
such as countries, scientific fields, or groups of institutions. Road travel time has been used
to explain co-patenting in Sweden (Andersson & Ejermo, 2005), with a subsequent study
using both road travel time and air travel time (Ejermo & Karlsson, 2006). Road travel
time has also been used when analyzing scholarly co-authorship networks in the Nether-
lands (Frenken, Hoekman, etal., 2009). Chinese authors have hypothesized that high-speed
railway accessibility may be one factor explaining the intensity of research collaboration
between cities in China (Ma etal., 2014). European data suggest that regions with major
international airports are likely to develop intensive international scientific collabora-
tion (Hoekman etal., 2010). A cross-sectional study of four large public US universities
found that more flight connections (connectivity) and closer airport proximity (accessibil-
ity) increase the expected number of co-authored scientific papers (Ploszaj etal., 2020).
Global flight network data have been used to analyze variation in citations of collaborative
research (Naik etal., 2023). A sample of scholars from the University of British Columbia
was used to investigate the influence of career stage, research productivity, field of exper-
tise, and other variables on academic air travel and the associated emissions (Wynes etal.,

Scientometrics
2019). Moreover, it has been found that universities with better connectivity via air trans-
port networks tend to be ranked higher (Guo etal., 2017). Among this research, several
papers stand out due to their use of a quasi-experimental approach. Catalini etal. (2020)
used a quasi-experimental design to examine the impact of a new, low-fare air route on
the probability of research collaboration. However, their research was limited in terms of
space and field, as they focused on United States-based chemistry scholars. Bahar etal.
(2023) analyzed relationships between direct long-haul flights and collaborative patent
applications. Dong etal. (2020) employed an instrumental variable approach to study the
impact of high-speed railways on scientific collaboration. Yao and Li (2022) and Kand
etal. (2023) analyzed to impact of Chinese high-speed rail on co-patenting. Similarly, Koh
etal. (2024) showed that the development of the Begin subway impacted collaborative pat-
ents on an intra-urban scale. Hu etal. (2022), using a difference-in-differences framework,
provide evidence that nonstop flights between China and the U.S. significantly enhance
the production of highly cited joint papers. However, most of these findings are limited to
single country, specifically US or China, restricting their broader applicability. Among the
above-mentioned quasi-experimental studies, only Bahar etal. (2023) examine the global
scale; however, their focus is on the impact of air availability on collaborative patents. In
contrast, the present study addresses a clear research gap by investigating the causal rela-
tionship between long-haul air accessibility and research collaboration, operationalized
through co-authored scientific publications.
Data andmethods
This study followed the natural experiment identified by Campante and Yanagizawa-
Drott (2018).They noted that cities that are just under 6000 miles apart are distinctly more
likely to have direct air links, especially compared with those slightly over the 6000-mile
threshold. They then determined that this effect did not correlate with any objective fac-
tors related to geography, geopolitics, economy, or aviation technology, but was rather due
to an arbitrary administrative decision. In accordance with aviation regulations in force
in many countries from the 1980s to the second decade of the twenty-first century, air-
lines were required to provide a double crew for flights scheduled to last longer than 12h,
including pilots and all other crew members. This 12-h limit was set arbitrarily. In prac-
tice, this meant that, for flights scheduled to last even 1min longer than 12h, the airline
had to pay for a double crew and provide a space where additional crew members could
rest. Therefore, exceeding the 12-h flight time limit resulted in a sudden increase in crew
costs. This cost factor presumably deterred airlines from launching such flights, an assump-
tion that was confirmed by Campante and Yanagizawa-Drott (2018). Using historical data,
they identified a clear discontinuity in the global air travel network exactly when expected:
at flight lengths of 12h. Furthermore, the authors noted that a 12-h flight translates to a
distance of roughly 6000 miles. This is an important observation because while data on
flight distance are readily available, data on flight duration are more difficult to obtain.
Substituting distance data for flight length data again shows the expected regularity. There
are significantly more connections between airport pairs just below 6000 miles than just
above this threshold. This jump is difficult to explain other than by the above-described
regulations. Since 12h to 6000 miles was arbitrarily determined, this situation can be used
as a natural experiment. Campante and Yanagizawa-Drott (2018) used this natural experi-
ment to investigate the causal relationship between long-haul air flights and economic

Scientometrics
development (using night lights satellite data as a proxy for economic development) and
business links (using international firm ownership data). In this paper, I applied Campante
and Yanagizawa-Drott’s (2018) natural experiment framework to other data, thereby not
only examining the phenomenon of scientific mobility and research collaboration but also
testing the natural experiment from a different angle.
An important aspect of this study’s methodology is its temporal dimension. The main
analyses were conducted for 2014. Far from being arbitrary, this choice related directly to
the type of natural experiment used. Air transport regulations that were the basis for the
discussed discontinuity have changed since 2014. New regulations have been introduced,
particularly in the United States (known as FAR 117) and the European Union (known as
EU FTL). As such, the 12-h (or 6000-mile) threshold became significantly less pronounced
after 2014. Consequently, 2014 is the last year for which the discussed natural experiment
can be realistically used for quasi-experimental analyses. On the other hand, before 2014,
the discontinuity had only occurred for 20-odd years. As Campante and Yanagizawa-Drott
(2018) showed for 1989, despite the same aviation regulations being in force, the 12-h limit
had no significant impact on the air transport network. This was because, prior to the early
1990s, there were a limited number of aircraft available that could cost-effectively fly such
long routes.
This paper used two main data sources. First, the International Civil Aviation Organiza-
tion’s proprietary datasets, “Traffic by Flight Stage” and “Airport Traffic,” which contain
data on individual air operations in international air traffic (not only on the completion of
the flight but also on the filling of seats available on a given flight, cargo volume, etc.), as
well as data enabling the characterization of international airports (number of operations,
number of passengers, etc.). The second source was Microsoft Academic Graph (MAG),
version 2020-09-01, as hosted by the Collaborative Archive & Data Research Environ-
ment project of the Indiana University Network Science Institute. Based on MAG data,
two dependent variables were calculated for this study (see below). The use of MAG as
a source of bibliometric data was motivated, firstly, by its open-access availability, and
secondly, by the fact that the affiliation data in MAG were geolocated, i.e., provided with
information on the longitude and latitude of a given institution (Sinha etal., 2015; Wang
etal., 2020). These two features clearly distinguished MAG from two other databases often
used in bibliometric analyses: the Web of Science and Scopus. Both these databases are
distributed on a commercial basis, and neither contain data on latitude and longitude.
Two dependent variables were used in this study. The first was defined as the number
of co-authored publications written by authors affiliated with institutions assigned to the
nearest international airport offering regular international flights (if there was more than
one airport in a given metropolitan area, their data were aggregated). In other words, the
variable indicates how many joint articles can be assigned to each pair of international
airports. For papers published in 2014, this variable had a mean of 15.5 and assumed a
value between 0 and 17,360. The latter value is the number of co-authored papers between
institutions assigned to the Boston airport and institutions assigned to airports serving New
York. Note that this variable was not normally distributed. The value of 0 (no co-authored
articles) applied to as many as 80% of cases, while 1–3 co-authored papers constituted
8.1% of the sample, 4–6 constituted 2.3%, 7–9 constituted 1.2%, and 10 or more consti-
tuted the remaining 8.4%.
The second dependent variable was defined as the percentage of international publica-
tions—publications with at least one author affiliated to an institution located in a differ-
ent country divided by the total number of publications at a given affiliation. This vari-
able was calculated for institutions spatially aggregated to spatial grid cells with a size of

Scientometrics
0.25 × 0.25°, which translates to a geographical space measuring approximately 27.5km
by 27.5km. Data aggregated at the grid-cell level allows for clear presentation on a map
(see Figs.1 and 2) and facilitates the modeling procedure using the distance variable of
the grid-cell centroid to the nearest airport. Since research activity is concentrated in a few
places on Earth, it is unsurprising that, for 98.1% of grid cells with a size of 0.25 × 0.25°,
the value of the variable is equal to 0. This is primarily due not to the absence of interna-
tional articles assigned to them but rather to the absence of affiliated articles in these loca-
tions. For grid cells with at least one affiliated article published in 2014, the variable ranges
from 0 to 100%, with an average of around 30%.
The study used two complementary quasi-experimental methods: regression discontinu-
ity (RD) design and instrumental variable (IV) design. Both methods build upon regres-
sion analysis, leveraging specific characteristics of the studied phenomenon to approximate
causal inference (Angrist & Pischke, 2009). A key aspect of this approach is the presence
Fig. 1 Share of internationally co-authored publications in grid cells of 0.25 × 0.25° for papers published in
2014
Fig. 2 Share of internationally co-authored publication in grid cells of 0.25 × 0.25° for papers published in
2014: close-up of selected areas

Scientometrics
of a naturally occurring condition that introduces an element of randomness, effectively
dividing the sample into a group exposed to a given influence (treatment group) and a
group that is not (control group). This structure mirrors a traditional experiment in which
random assignment underpins causal inference. In this analysis, the critical characteristic is
the division of airport pairs into those just below and just above the 6000-mile threshold.
As demonstrated above, this arbitrarily defined threshold significantly affects the probabil-
ity of direct air connections between airports. By exploiting this exogenous variation, this
analytical framework enables a robust examination of how long-haul air accessibility influ-
ences research collaboration across distant locations. The use of both RD and IV in this
study allowed the use of two dependent variables, complementary, but with very different
characteristics: network-pair data in the case of RD and grid-cell data in the case of IV. RD
design seems more straightforward in its interpretation—especially since the result can be
presented using striking and easy-to-read charts showing how values change in relation to
the threshold. On the other hand, the use of instrumental variable design facilitated com-
plementary analyses showing the spatial extent of the airport’s impact.
The specification of the models largely followed Campate and Yanagizawa-Drott’s
(2018) proposed framework. The RD model was formulated as follows:
where Yij denotes the dependent variable, defined as the number of publications by co-
authors affiliated with institutions located in places assigned to individual “airport pairs.”
Below6Kij is a discontinuity dummy equal to 1 if dij is less than 6000 miles. The proposed
specification allowed the inclusion of control variables and various slopes above and below
the 6000-mile threshold, as well as to model nonlinearities (by inclusion of polynomials).
In turn, the instrumental variable model was formulated as follows:
where Yic denotes the dependent variable of the percentage of publications written in inter-
national collaboration, defined at the grid-cell level (see Figs.1 and 2). ShareEIBelow6Kic
is a composite instrumental variable invented and calculated by Campante and Yanagi-
zawa-Drott (2018), which they defined as the sum of the eigenvector centrality of airports
5,500 to 6000 miles away from airport i, divided by the sum of the eigenvector centrality
of airports 5,500 to 6,500 miles away. The use of eigenvector centrality in the calculation
of the variable made it possible to take into account not only the direct availability of long-
haul flights but also the availability of connecting flights.
Finally, to account for the heterogeneity of the studied places and airports in the esti-
mated models, a number of control variables were used, again in accordance with Cam-
pante and Yanagizawa-Drott (2018). The following control variables were included: total
number of flights; number of daily, twice weekly, and weekly flights; total number of pas-
sengers and seats; number of connected cities and countries; time zone; distance to the
equator; and GDP per capita at the country level. In the below model specifications, these
variables are referred to as “airport controls.” In addition, this study introduced a control
variable specific to its context: “scientific capacity”. In IV models, it is defined as the num-
ber of publications assigned to a given location, while in RD models, it is measured as the
product of the number of publications assigned to airport areas i and j, Adding control vari-
ables to the RD model requires that they satisfy the covariate continuity assumption, mean-
ing there should be no discontinuities in the covariates at the threshold of the assignment
variable. The fulfillment of this assumption for airport controls was verified by Campante
Yij
=𝛼+𝛽∗Below6K
ij
+g
(
d
ij)
∗𝛾+𝜀
ij,
Yic
=
𝛼
+
𝛽
∗
ShareEIBelow6Kic
+
Xic𝛾
+
𝜀ic,

Scientometrics
and Yanagizawa-Drott (2018). For the scientific capacity variable, the absence of disconti-
nuities at the key cutoff point is illustrated in a graph included in the online appendix.
Results
Figure3 illustrates the main results of this analysis and shows the number of co-authored
publications by distance between airports. For the sake of readability, the number of arti-
cles has been aggregated into bins covering 100 miles each. The vertical line in the mid-
dle of the chart denotes the presumed cutoff threshold of 6000 miles. The results clearly
demonstrate the significance of this threshold. Not only does exceeding the 6000-mile limit
result in a significant drop in the number of co-authored publications, but the trend lines on
both sides of the cutoff threshold are completely different. As such, the figure provides a
strong basis for expecting that the anticipated effect (hypothesis of the paper) is supported
by the data. The modeling results further confirm this expectation.
All six specifications of the RD model produced statistically significant results
(Table1). Depending on the specification, exceeding the 6000-mile limit resulted in reduc-
tions of between 2.6 and eight in the expected number of co-authored publications. Models
(1) and (2) used an arbitrarily selected bandwidth (i.e., the distance to the right and left of
the cutoff threshold), which determined the data included in the model. This was 500 miles
in Model (1) and 1000 miles in Model (2). Subsequent models used the so-called optimal
bandwidth (i.e., the bandwidth determined algorithmically by the model implemented in
STATA based on given data characteristics). The use of alternative bandwidths is a good
practice in RD modeling and a method of ensuring the robustness of the results (Imbens
& Lemieux, 2008). In this case, the first three models differing only in bandwidth width
generated similar RD estimate values and standard errors, which significantly increased the
reliability of the obtained results. The remaining specifications (4–6) added control vari-
ables characterizing the studied airports to the model. Specifications (5) and (6) added a
Fig. 3 Number of co-authored publications by distance between closest airports in 2014

Scientometrics
scientific capacity variable defined as the product of the number of publications assigned
to airport areas i and j. The addition of the scientific capacity variable is consistent with the
gravity model theoretical approach. The last specification (6) introduced a second-order
polynomial to enable the modeling of curvilinear trends on both sides of the cutoff thresh-
old. It should be emphasized that adding the scientific capacity variable clearly reduced the
RD estimate value to −2.6 in the case of specification (5) and −3.2 in the case of specifi-
cation (6), while still remaining statistically significant.
As explained above, the 6000-mile discontinuity became visible in the 1990s. In 1998,
this discontinuity appeared practically nonexistent. Therefore, it could be expected that the
effect observed in the 2014 data would be absent in 1989. The collected data confirmed
this intuition to a certain extent. Figure4 shows that, while there is a decline in the number
of co-authored papers after crossing the 6000-mile threshold, it is also clear that this is part
of a broader trend, wherein the number of co-authored publications declines with increas-
ing distance. If there is a discontinuity here, it is certainly less pronounced compared to the
case of the 2014 data.
The modeling results confirm the conclusions drawn from the observations in Fig.2.
Specifications (1–4), those that do not account for the scientific capacity variable, provide
a statistically significant result, indicating the influence of the 6000-mile discontinuity.
However, the full specifications (5) and (6), which do account for the scientific capacity
(according to the gravity model) deliver a statistically insignificant result (see Table 2).
This result aligns with expectations and ultimately strengthens the conclusion of the overall
analysis. Since there was no discontinuity in the air traffic network at the 6000-mile point
in 1989, it should not affect other changes. However, in 2014 the discontinuity at the same
point was fully developed and could thus affect other variables—in our case, co-authored
publications.
The second part of the analysis using the instrumental variable framework largely
confirmed the results of the RD analysis. A positive estimate of the instrumental var-
iable would indicate that, the better the developed network of long-distance connec-
tions, the greater the percentage of international co-authored publications (see Table3).
According to the assumptions of the gravity model, scientific capacity, calculated as
the total number of scientific publications assigned to a given grid cell, was ultimately
Table 1 Effect on the number of co-authored publications, airport-pair level, in 2014
Significance levels: ***p < 0.01, **p < 0.05, *p < 0.1
Dependent variable: number of co-authored publications
Specification (1) (2) (3) (4) (5) (6)
Bandwidth in miles 500 miles 1000 miles Optimal Optimal Optimal Optimal
RD estimate −6.13 −8.03 −7.98 −6.47 −2.57 −3.24
Robust standard errors (2.58)** (1.75)*** (1.77)*** (1.51)*** (1.10)** (1.33)**
Covariates:
Airport controls ✔ ✔ ✔
Scientific capacity ✔ ✔
Polynomial order 1st 1st 1st 1st 1st 2nd
Observations: total 334,954 334,954 334,954 334,954 334,954 334,954
Observations: effective 41,477 81,842 80,540 87,562 86,027 124,884

Scientometrics
a statistically significant variable. It is worth noting that adding baseline data on the
percentage of international publications in 1989 (i.e., before the discontinuity occurred)
to the model significantly increases its ability to explain the variance of the dependent
variable. The R-squared value in specification (2) in Table3 is 0.27, which is signifi-
cantly higher than in the case of specification (1), which does not include this baseline
variable.
The specifications presented in Table4 show the extent of international airports’
influence on international research collaboration. Specification (1) in Table4 uses the
interaction between the instrumental and distance variables of a given grid cell to the
nearest international airport within the same country. A negative estimate means that, as
Fig. 4 Number of co-authored publications by distance between closest airports in 1989
Table 2 Effect on the number of co-authored publications, airport-pair level, in 1989
Significance levels: ***p < 0.01, **p < 0.05, *p < 0.1
Dependent variable: number of co-authored publications
Specification (1) (2) (3) (4) (5) (6)
Bandwidth 500 miles 1000 miles Optimal Optimal Optimal Optimal
RD estimate −0.065 −0.078 −0.074 −0.060 −0.014 −0.018
Robust standard errors (0.027)** (0.020)*** (0.021)*** (0.019)*** (0.012) (0.016)
Covariates:
Airport controls ✔ ✔ ✔
Scientific capacity ✔ ✔
Polynomial order 1st 1st 1st 1st 1st 2nd
Observations: total 334,954 334,954 334,954 334,954 334,954 334,954
Observations: effective 41,477 81,842 70,745 76,155 109,765 109,765

Scientometrics
the distance increases, the airport’s positive impact gradually decreases. In turn, specifi-
cations (2–7) were calculated for subsamples determined according to the distance from
the airport: grid cells located up to 50 miles from the airport, then located at a distance
from 50 to 100 miles, then located at a distance of 100 miles to 150 miles, etc., with an
increment of 50 miles up to a value of 300 miles. The obtained results clearly show that
the influence of the airport is highest in its vicinity. The estimate of the instrumental
variable for the subsample below 50 miles was three times higher than for the 50–100
and 100–150-mile subsamples. Moreover, the result for the subsample below 50 miles
was more statistically significant than those for the next two intervals. After crossing the
150-mile limit, the instrumental variable was no longer statistically significant, meaning
that the analyzed impact reached no farther than 150 miles from the airport.
Discussion andconclusions
The analyses presented in this article support the hypothesis that greater availability of
long-distance air connections positively affects long-distance research collaboration. The
effect was observed and statistically significant for both tested dependent variables and
both quasi-experimental methods. The research results are consistent with the findings
of previous works examining the relationship between air accessibility and scientific col-
laboration (Andersson & Ejermo, 2005; Catalini etal., 2020; Hoekman etal., 2010; Plo-
szaj etal., 2020) or, more broadly, between transport accessibility—also accounting for
additional means of transport—and scientific collaboration (Dong etal., 2020; Ejermo &
Karlsson, 2006; Ma etal., 2014). Against this background, the main contribution of this
paper is the application of a quasi-experimental method to global data. This is a signifi-
cant extension of the only paper to date to have employed a quasi-experimental approach
to study the relationship between air accessibility and scientific collaboration—a paper
examining the impact of the introduction of low-cost airline flights on collaboration among
chemistry scholars in the United States (Catalini etal., 2020).
The availability of long-distance air connections is neither the main nor the decisive
factor in initiating and conducting fruitful research collaboration (Hall etal., 2018; Katz
& Martin, 1997; Leahey, 2016; Sonnenwald, 2007). In the broadest sense, an appropriate
Table 3 Effect on the share
of internationally co-authored
publications, grid-cell level, in
2014
Significance levels: ***p < 0.01, **p < 0.05, *p < 0.1
Dependent variable: share of internationally co-authored publications
(%)
Specification (1) (2)
ShareEI_below6000_w500 0.15*** 0.12***
(0.05) (0.04)
Scientific capacity 0.19*** 0.13***
(0.04) (0.03)
Co-authored publications in 1989 0.36***
(0.03)
Airport controls ✔ ✔
Observations 37,812 37,812
R-squared 0.15 0.27

Scientometrics
Table 4 Effect on the share of internationally co-authored publications, spatial patterns, grid-cell level, in 2014
Significance levels: ***p < 0.01, **p < 0.05, *p < 0.1
Dependent variable: share of internationally co-authored publications (%)
Specification (1) (2) (3) (4) (5) (6) (7)
Subsample: miles from closest airport < 500 < 50 50–100 100–150 150–200 200–250 250–300
ShareEI_below6000_w500 1.28*** 0.45*** 0.15** 0.16** 0.03 0.01 0.01
(0.46) (0.15) (0.07) (0.07) (0.04) (0.03) (0.02)
ShareEI_below6000 *Distance to airport (interaction) −0.23***
(0.09)
Covariates: airport controls, scientific capacity, co-authored
publications in 1989
✔ ✔ ✔ ✔ ✔ ✔ ✔
Observations 157,341 14,480 23,332 23,206 20,717 17,615 15,647
R-squared 0.23 0.31 0.20 0.17 0.11 0.15 0.27

Scientometrics
capacity for research collaboration is needed. The results presented in this paper confirm
this assumption. Model specifications that account for the scientific potential of collabo-
rators (conceptualized here simply as the number of publications) explain a greater part
of the variability of the outcome variables than models that do not take this variable into
account. This is also consistent with, and expected on the basis of, the gravity model the-
ory. Moreover, in models that account for scientific capacity, the estimates of the impact
of air accessibility are lower and/or less statistically significant than in the case of models
that neglect this variable. However, it should be emphasized that air accessibility variables
remain statistically significant.
The results clearly show that the positive impact of an international airport offering
long-distance connections is spatially limited: the impact is greatest near the airport, grad-
ually decreases with increasing distance, and becomes negligible once the distance exceeds
150 miles. This result is consistent with the results of previous (non-quasi-experimental)
studies (Ploszaj etal., 2020).
In a broad sense, this study fits into the discussion on the relationship between trans-
port infrastructure and spatial connections, or socioeconomic development in general
(Crescenzi & Rodríguez‐Pose, 2012). Views on this issue range between two extremes. On
the one hand, it is said that transport infrastructure influences the emergence of flows and
connections. On the other, it has been argued that infrastructure should be created where
these connections already exist. The results presented in this paper provide evidence that
transport infrastructure can translate into increased collaboration, even over significant dis-
tances. At the same time, it should be emphasized that these results do not mean that there
is no reverse relationship, that is, that existing relationships (including scientific collabo-
ration) stimulate the launch of new air connections. Demand is, of course, an important
factor in the supply of air services (Wang & Gao, 2021). However, this issue goes beyond
the scope of this study and is impossible to investigate using the natural experiment imple-
mented here.
An important limitation of this study is its inherently historical nature. Two aspects are
worth mentioning here. Firstly, due to the abovementioned changes in aviation regulations,
discontinuity (i.e., the basis for the natural experiment) declined after 2014. Accordingly,
it would be impossible to repeat these analyses to obtain more current data. Secondly,
the circumstances of scientific collaboration have changed drastically as a result of the
COVID-19 pandemic. Remote collaboration tools have improved significantly and become
common, causing, among other things, the (perhaps temporary) spread of virtual scientific
conferences (Falk & Hagsten, 2021; Olechnicka etal., 2024, 2025). It is difficult to predict
the extent to which virtual contacts will replace the need for scientific mobility. However,
it should certainly be stated that new research on the relationship between air accessibil-
ity and scientific collaboration using data from the new, post-pandemic reality, is neces-
sary. Moreover, such research should also endeavor to include virtual communication as an
important covariate.
Future analyses of the impact of transport accessibility on research collaboration should
consider the potential crowding-out effect, where the development of one type of collabo-
ration may come at the expense of another. A study by Hu etal. (2022) examines whether
increased international nonstop flights affect domestic and non-U.S. collaborations in
China. Their findings suggest that the introduction of U.S.-China nonstop flights led to a
shift in collaboration patterns, although the evidence for a strong crowding-out effect was
inconclusive. Including this perspective in future studies could provide a more comprehen-
sive understanding of whether improved long-haul air accessibility reallocates rather than
expands collaboration networks, potentially affecting local or regional research dynamics.

Scientometrics
Also, the issue of science policy implications ought to be considered. If direct long-dis-
tance air connections are beneficial for the development of research collaboration, should
promoting the creation of new air connections be an element of scientific policy? This
conclusion seems too far-reaching. Let us underline that better transport accessibility can
only facilitate academic mobility, but it is not a sufficient condition for the development of
research collaboration. Furthermore, while the role of physical mobility in research col-
laboration is well-established, recent geopolitical developments highlight additional con-
straints beyond infrastructure limitations. For example, Wang etal. (2023) demonstrate that
political tensions between the U.S. and China have led to a measurable decline in air pas-
senger flows, particularly affecting travel to university hubs, thereby potentially influencing
international academic exchange.
Another important aspect emerging from the recommendations of this study is the
harmful impact of air transport on the climate. Air transport emits relatively large amounts
of greenhouse gases and scientists are one of the most internationally mobile professional
groups (Arsenault etal., 2019; Hölbling etal., 2023; Schmidt, 2022). The challenge for
science policy, at both the national and institutional levels, is therefore to find a balance
between supporting the mobility of scientists and promoting environmental responsibility.
On this basis, it could be argued that future research on the mobility of scientists, and its
causes and effects, should focus on assessing the effectiveness of different types of mobil-
ity (e.g., short vs. long term) and the differences in effectiveness related to the character-
istics of scientists (e.g., early vs. late career stages). Such efficiency assessments should
consider environmental costs. Moreover, future research could also assess which forms of
academic mobility could be replaced by remote communication, to what extent, and under
what conditions.
Supplementary Information The online version contains supplementary material available at https:// doi.
org/ 10. 1007/ s11192- 025- 05291-5.
Acknowledgements This work was supported by the National Science Centre Poland under Grant
2018/31/B/HS4/03997. Data for the project were made available by the Indiana University Network Sci-
ence Institute and the Science Studies Laboratory of the University of Warsaw. I would like to thank Katy
Börner for her warm hospitality and kind support during my stay at the Cyberinfrastructure for Network
Science (CNS) Center at Indiana University Bloomington. I thank David Yanagizawa-Drott for kindly pro-
viding source data and code. I thank Xiaoran Yan for his generous assistance in processing bibliometric
data. Finally, I would like to thank two anonymous reviewers whose insightful comments were very helpful
in developing the final version of the paper.
Declarations
Conflict of interest The author has no competing interests to declare that are relevant to the content of this
article.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License,
which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long
as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Com-
mons licence, and indicate if changes were made. The images or other third party material in this article
are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the
material. If material is not included in the article’s Creative Commons licence and your intended use is not
permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly
from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.

Scientometrics
References
Allen, T. J. (2007). Managing the flow of technology: Technology transfer and the dissemination of techno-
logical information within the R&D organization. MIT Press.
Andersson, Å. E., & Persson, O. (1993). Networking scientists. The Annals of Regional Science, 27(1),
11–21. https:// doi. org/ 10. 1007/ BF015 81830
Andersson, M., & Ejermo, O. (2005). How does accessibility to knowledge sources affect the innovative-
ness of corporations?—Evidence from Sweden. The Annals of Regional Science, 39(4), 741–765.
https:// doi. org/ 10. 1007/ s00168- 005- 0025-7
Angrist, J., & Pischke, J.-S. (2009). Mostly harmless econometrics. Princeton University Press.
Arndt, R. T. (1996). The fulbright difference. Transaction Publ.
Arsenault, J., Talbot, J., Boustani, L., Gonzalès, R., & Manaugh, K. (2019). The environmental footprint of
academic and student mobility in a large research-oriented university. Environmental Research Letters,
14(9), 095001. https:// doi. org/ 10. 1088/ 1748- 9326/ ab33e6
Bahar, D., Choudhury, P., Kim, D. Y., & Koo, W. W. (2023). Innovation on wings: Nonstop flights and firm
innovation in the global context. Management Science, 69(10), 6202–6223. https:// doi. org/ 10. 1287/
mnsc. 2023. 4682
Boudreau, K. J., Brady, T., Ganguli, I., Gaule, P., Guinan, E., Hollenberg, A., & Lakhani, K. R. (2017). A
field experiment on search costs and the formation of scientific collaborations. The Review of Econom-
ics and Statistics, 99(4), 565–576. https:// doi. org/ 10. 1162/ REST_a_ 00676
Campante, F., & Yanagizawa-Drott, D. (2018). Long-range growth: Economic development in the global
network of air links*. The Quarterly Journal of Economics, 133(3), 1395–1458. https:// doi. org/ 10.
1093/ qje/ qjx050
Campbell, D. T., & Stanley, J. C. (2011). Experimental and quasi-experimental designs for research.
Wadsworth.
Catalini, C., Fons-Rosen, C., & Gaulé, P. (2020). How do travel costs shape collaboration? Management
Science, 66(8), 3340–3360. https:// doi. org/ 10. 1287/ mnsc. 2019. 3381
Chai, S., & Freeman, R. B. (2019). Temporary colocation and collaborative discovery: Who confers at con-
ferences. Strategic Management Journal, 40(13), 2138–2164. https:// doi. org/ 10. 1002/ smj. 3062
Crescenzi, R., & Rodríguez-Pose, A. (2012). Infrastructure and regional growth in the European Union.
Papers in Regional Science, 91(3), 487–513. https:// doi. org/ 10. 1111/j. 1435- 5957. 2012. 00439.x
Dong, X., Zheng, S., & Kahn, M. E. (2020). The role of transportation speed in facilitating high skilled
teamwork across cities. Journal of Urban Economics, 115, 103212. https:// doi. org/ 10. 1016/j. jue. 2019.
103212
Duede, E., Teplitskiy, M., Lakhani, K., & Evans, J. (2024). Being together in place as a catalyst for scientific
advance. Research Policy, 53(2), 104911. https:// doi. org/ 10. 1016/j. respol. 2023. 104911
Dunning, T. (2012). Natural experiments in the social sciences: A design-based approach. Cambridge Uni-
versity Press.
Ejermo, O., & Karlsson, C. (2006). Interregional inventor networks as studied by patent coinventorships.
Research Policy, 35(3), 412–430. https:// doi. org/ 10. 1016/j. respol. 2006. 01. 001
Falk, M. T., & Hagsten, E. (2021). When international academic conferences go virtual. Scientometrics,
126(1), 707–724. https:// doi. org/ 10. 1007/ s11192- 020- 03754-5
Fernández, A., Ferrándiz, E., & León, M. D. (2016). Proximity dimensions and scientific collaboration
among academic institutions in Europe: The closer, the better? Scientometrics, 106(3), 1073–1092.
https:// doi. org/ 10. 1007/ s11192- 015- 1819-8
Frenken, K., Hardeman, S., & Hoekman, J. (2009a). Spatial scientometrics: Towards a cumulative research
program. Journal of Informetrics, 3(3), 222–232. https:// doi. org/ 10. 1016/j. joi. 2009. 03. 005
Frenken, K., Hoekman, J., Kok, S., Ponds, R., van Oort, F., & van Vliet, J. (2009b). Death of distance in
science? A gravity approach to research collaboration. In A. Pyka & A. Scharnhorst (Eds.), Innovation
Networks (pp. 43–57). Springer.
Guo, W., Del Vecchio, M., & Pogrebna, G. (2017). Global network centrality of university rankings. Royal
Society Open Science, 4(10), 171172. https:// doi. org/ 10. 1098/ rsos. 171172
Hall, K. L., Vogel, A. L., Huang, G. C., Serrano, K. J., Rice, E. L., Tsakraklides, S. P., & Fiore, S. M.
(2018). The science of team science: A review of the empirical evidence and research gaps on collabo-
ration in science. American Psychologist, 73(4), 532–548. https:// doi. org/ 10. 1037/ amp00 00319
Hoekman, J., Frenken, K., & Tijssen, R. J. W. (2010). Research collaboration at a distance: Changing spatial
patterns of scientific collaboration within Europe. Research Policy, 39(5), 662–673. https:// doi. org/ 10.
1016/j. respol. 2010. 01. 012
Hoekman, J., Frenken, K., & van Oort, F. (2009). The geography of collaborative knowledge production in
Europe. The Annals of Regional Science, 43(3), 721–738. https:// doi. org/ 10. 1007/ s00168- 008- 0252-9

Scientometrics
Hölbling, S., Kirchengast, G., & Danzer, J. (2023). Unmasking mobility patterns: International travel behav-
ior and emissions of scientists in a higher research institution. International Journal of Sustainability
in Higher Education, 24(9), 355–371. https:// doi. org/ 10. 1108/ IJSHE- 03- 2023- 0081
Hu, G., Ni, R., & Tang, L. (2022). Do international nonstop flights foster influential research? Evidence
from Sino-US scientific collaboration. Journal of Informetrics, 16(4), 101348. https:// doi. org/ 10.
1016/j. joi. 2022. 101348
Imbens, G. W., & Lemieux, T. (2008). Regression discontinuity designs: A guide to practice. Journal of
Econometrics, 142(2), 615–635. https:// doi. org/ 10. 1016/j. jecon om. 2007. 05. 001
Jacob, M., & Meek, V. L. (2013). Scientific mobility and international research networks: Trends and
policy tools for promoting research excellence and capacity building. Studies in Higher Education,
38(3), 331–344. https:// doi. org/ 10. 1080/ 03075 079. 2013. 773789
Jin, Z., Cui, Z., Zhu, S., & Xu, L. (2024). Face-to-face contact and university–industry collaboration:
Evidence from mobile signaling data in Beijing. The Annals of Regional Science, 72(4), 1255–
1276. https:// doi. org/ 10. 1007/ s00168- 023- 01242-1
Kabo, F. W., Cotton-Nessler, N., Hwang, Y., Levenstein, M. C., & Owen-Smith, J. (2014). Proximity
effects on the dynamics and outcomes of scientific collaborations. Research Policy, 43(9), 1469–
1485. https:// doi. org/ 10. 1016/j. respol. 2014. 04. 007
Kang, M., Li, Y., Zhao, Z., Song, M., & Yi, J. (2023). Travel costs and inter-city collaborative innova-
tion: Evidence of high-speed railway in China. Structural Change and Economic Dynamics, 65,
286–302. https:// doi. org/ 10. 1016/j. strue co. 2023. 02. 002
Katz, J. S., & Martin, B. R. (1997). What is research collaboration? Research Policy, 26(1), 1–18.
https:// doi. org/ 10. 1016/ S0048- 7333(96) 00917-1
Koh, Y., Li, J., & Xu, J. (2024). Subway, collaborative matching, and innovation. Review of Economics
and Statistics. https:// doi. org/ 10. 1162/ rest_a_ 01279
Leahey, E. (2016). From sole investigator to team scientist: Trends in the practice and study of
research collaboration. Annual Review of Sociology, 42(1), 81–100. https:// doi. org/ 10. 1146/ annur
ev- soc- 081715- 074219
Ma, H., Fang, C., Pang, B., & Li, G. (2014). The effect of geographical proximity on scientific coopera-
tion among Chinese cities from 1990 to 2010. PLoS ONE, 9(11), e111705. https:// doi. org/ 10. 1371/
journ al. pone. 01117 05
Naik, C., Sugimoto, C. R., Larivière, V., Leng, C., & Guo, W. (2023). Impact of geographic diversity on
citation of collaborative research. Quantitative Science Studies, 4(2), 442–465. https:// doi. org/ 10.
1162/ qss_a_ 00248
Netz, N., Hampel, S., & Aman, V. (2020). What effects does international mobility have on scientists’
careers? A Systematic Review. Research Evaluation, 29(3), 327–351. https:// doi. org/ 10. 1093/ resev
al/ rvaa0 07
Olechnicka, A., Ploszaj, A., & Celinska-Janowicz, D. (2019). The geography of scientific collaboration.
Routledge.
Olechnicka, A., Ploszaj, A., & Zegler-Poleska, E. (2024). Virtual academic conferencing: A scoping
review of 1984–2021 literature. Novel modalities vs. long standing challenges in scholarly commu-
nication. Iberoamerican Journal of Science Measurement and Communication, 4(1), 1–31. https://
doi. org/ 10. 47909/ ijsmc. 93
Olechnicka, A., Ploszaj, A., & Zegler-Poleska, E. (2025). The impact of the virtualization of scholarly
conferences on the gender structure of conference contributors. Scientometrics, 130(1), 423–445.
https:// doi. org/ 10. 1007/ s11192- 024- 05212-y
Petersen, A. M. (2018). Multiscale impact of researcher mobility. Journal of the Royal Society Interface,
15(146), 20180580. https:// doi. org/ 10. 1098/ rsif. 2018. 0580
Picci, L. (2010). The internationalization of inventive activity: A gravity model using patent data.
Research Policy, 39(8), 1070–1081. https:// doi. org/ 10. 1016/j. respol. 2010. 05. 007
Ploszaj, A., Yan, X., & Börner, K. (2020). The impact of air transport availability on research collabora-
tion: A case study of four universities. PLoS ONE, 15(9), e0238360. https:// doi. org/ 10. 1371/ journ
al. pone. 02383 60
Plotnikova, T., & Rake, B. (2014). Collaboration in pharmaceutical research: Exploration of country-
level determinants. Scientometrics, 98(2), 1173–1202. https:// doi. org/ 10. 1007/ s11192- 013- 1182-6
Schmidt, A. (2022). University air travel and greenhouse gas mitigation: An analysis of higher education
climate policies. International Journal of Sustainability in Higher Education, 23(6), 1426–1442.
https:// doi. org/ 10. 1108/ IJSHE- 07- 2021- 0318
Sebestyén, T., & Varga, A. (2013). Research productivity and the quality of interregional knowl-
edge networks. The Annals of Regional Science, 51(1), 155–189. https:// doi. org/ 10. 1007/
s00168- 012- 0545-x

Scientometrics
Sekhon, J. S., & Titiunik, R. (2012). When natural experiments are neither natural nor experiments.
American Political Science Review, 106(1), 35–57. https:// doi. org/ 10. 1017/ S0003 05541 10005 42
Sinha, A., Shen, Z., Song, Y., Ma, H., Eide, D., Hsu, B.-J. (Paul), & Wang, K. (2015). An overview of
microsoft academic service (MAS) and applications. In: Proceedings of the 24th international con-
ference on world wide web, 243–246. https:// doi. org/ 10. 1145/ 27409 08. 27428 39
Sonnenwald, D. H. (2007). Scientific collaboration. Annual Review of Information Science and Technol-
ogy, 41(1), 643–681. https:// doi. org/ 10. 1002/ aris. 2007. 14404 10121
Souto-Otero, M, Humburg, M, Franke, J, & European Commission, Directorate-General for Education,
Youth, Sport and Culture. (2017). FP7 ex post and H2020 interim evaluation of Marie Skłodowska-
Curie actions (MSCA): Final report. Publications Office. https:// doi. org/ 10. 2766/ 103174
Wang, K., Shen, Z., Huang, C., Wu, C.-H., Dong, Y., & Kanakia, A. (2020). Microsoft Academic Graph:
When experts are not enough. Quantitative Science Studies, 1(1), 396–413. https:// doi. org/ 10. 1162/
qss_a_ 00021
Wang, S., & Gao, Y. (2021). A literature review and citation analyses of air travel demand studies published
between 2010 and 2020. Journal of Air Transport Management, 97, 102135. https:// doi. org/ 10. 1016/j.
jairt raman. 2021. 102135
Wang, Z., Tang, L., Cao, C., & Zhou, Z. (2023). The impact of U.S.-China tensions on people mobility.
China Review, 23(4), 159–195.
Wynes, S., Donner, S. D., Tannason, S., & Nabors, N. (2019). Academic air travel has a limited influence
on professional success. Journal of Cleaner Production, 226, 959–967. https:// doi. org/ 10. 1016/j. jclep
ro. 2019. 04. 109
Yao, L., & Li, J. (2022). Intercity innovation collaboration and the role of high-speed rail connections: Evi-
dence from Chinese co-patent data. Regional Studies, 56(11), 1845–1857. https:// doi. org/ 10. 1080/
00343 404. 2021. 20083 40
Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and
institutional affiliations.