The structure and dynamics of instrument collaboration networks

Abstract

Complex scientific questions often require collaboration between scientists to access scientific instruments (deS. Price, Res Policy 13:3–20, 1984; Shrum et al. 2007, Structures of scientific collaboration, The MIT Press, 2007), knowledge and social capital from scientists outside of their immediate networks (Burt, Am J Sociol 110:349–399, 2004; Collins, Tacit and explicit knowledge, University of Chicago Press, https://press.uchicago.edu/ucp/books/book/chicago/T/bo8461024.html, 2012; Granovetter, Am J Sociol 78:1360–1380, 1973; Polanyi, Personal knowledge: towards a post-critical philosophy (Repr. (with corr.)). Routledge & Kegan Paul, 1962). Synchrotron radiation facilities are a unique case to explore this type of collaboration, as external scientists going to the facility to do their ordinary research work are common. These external users use complex scientific instrumentation that could require the additional accumulated knowledge of internal scientists and staff more familiar with the technical and social aspects of the instruments. These collaborations sometimes result in a co-authored publication with internal staff, employed in these facilities. However, this is not always the case. Data from the European Synchrotron Radiation Facility (ESRF) shows that most often, external scientists do not include internal staff in the co-author fields of the subsequent publication from scientific work done in one (or more) instruments in these facilities. Instrument collaboration networks are constructed and analysed over the period 2000–2018 for the different scientific instruments within the facility. A strong relationship between the level of external collaboration and the structure of the networks is found and explored. The results provide further insight into factors that shape collaboration and knowledge transfer, also relevant to policy makers and facility managers seeking to promote these activities.

Introduction

Cutting edge scientific instrumentation, which tend to be expensive, complicated to use, and/or with limited availability, along with their experimental techniques, are often crucial for developing scientific ideas and discoveries into the practical, applied and tangible (deS. Price, 1984). One reason scientists collaborate is to access this type of technology—defined as the set of instruments and practices scientists employ in the acquisition and manipulation of information (Shrum et al., 2007), as well as the arguably tacit knowledge arguably (Collins, 2012; Polanyi, 1962) and social capital from scientists in critical positions in a network (Burt, 2004; Granovetter, 1973).

With co-authorship as a proxy for collaboration, a wide range of studies have been providing windows on the patterns of collaboration in semi-fixed academic disciplines or communities (Bettencourt et al., 2009; Fonseca et al., 2016; Kumar, 2015; Mali et al., 2012; Newman, 2001, 2004b, 2004a). However, collaboration can and often does transcend traditional institutional or disciplinary boundaries and the role of technology and specifically, scientific instruments remains an important but understudied phenomenon in studies of science (deS. Price, 1984; Hallonsten & Heinze, 2015; Heinze, 2013; Heinze et al., 2013; Mody, 2011).

Thus, a theoretical framework based on de Solla Price (1984) Granovetter (1973), Burt (2004) and Shrum et al. (2007), shifts the focus of the collaboration network to the scientific instrument to deepen the understanding of collaboration between scientists in the backdrop of synchrotron radiation facilities, where collaboration between internal and external scientist exist. Data gathered for the European Synchrotron Radiation Facility identifies these links. With it, the study aims to examine the collaboration networks between different scientific instruments, with a focus on the relationship between the level of external use of these instruments and the structure of the networks. The study also explores how the networks between the different scientific instruments are formed and change over time and proposes some reasons behind the different network structures.

The results of the study suggest that there is a significant relationship between these factors, as instruments that have a higher level of external use are associated with more fragmented collaboration networks compared to instruments with lower levels of external use. One reason behind the differences might be related to technological advancements to the instruments that increase their ease-of-use for external scientists, and how that affects the willingness of formalizing co-authorship with internal scientists and staff. These findings provide insight into the factors that shape collaboration networks, and the ways that different teams interact and share knowledge. Furthermore, it can provide insight into which conditions are necessary to formally recognize co-authorship in scientific work. This can inform research on collaboration and knowledge transfer in science and technology and help identify strategies for promoting collaboration and facilitating the exchange of expertise and knowledge.

The article is divided as follows. Section 2 presents the “Previous and related research”, Section 3 presents the “Material and methods”, and Section 4 presents the “Theoretical framework and motivation for the instrument view”. Section 5 presents the “Results”, with Sections 5 and 6 providing a “Discussion” and “Conclusion” of the study.

Previous and related research

Co-authorship is only one dimension of scientific collaboration, which has different forms arising from social conventions among scientists with fuzzy lines between the less formal links among scientists and the formal collaboration often seen in co-authorship. Not all collaboration is shown in the co-authorship level and not all co-authors strictly collaborate with each other (Katz & Martin, 1997). Some early work of scientific collaboration include the theorization of collaboration via the professionalization of science from a sociological and historical perspective (Beaver & Rosen, 1978, 1979a, 1979b); the analysis of the determinants of international collaboration (Luukkonen et al., 1992); the structure and change of collaborative networks with statistical indicators (Melin & Persson, 1996); and the analysis of structural changes in international co-authorship links (Glanzel, 2001).

Collaboration networks of the kind used in this study aim to model some social network where nodes are connected due to a collaborative relationship of some kind. The scientific collaboration network is explicit in understanding relationships between scientists based on paper co-authorship. In its simplest definition, two scientists are considered connected if they authored a paper together (Newman, 2001). Co-authorship networks have served as a prototype for studying complex evolving networks from a social network perspective (Barabasi et al., 2002); as a tool to understand the growth of international co-authorship (Wagner & Leydesdorff, 2005); and to analyse the evolving structure of collaboration networks within scientific disciplines (Bettencourt et al., 2009) and research institutes (di Bella et al., 2021). However, they have been mostly used to understand the relationships formed within disciplines, journals, or institutions.

Related research on collaboration or co-authorship networks from synchrotron radiation facilities is scarce: a brief discussion of collaboration between in-house (internal) scientists and external users (Hallonsten, 2016a) and an analysis of the Swiss Light Source by Silva et al. (2019), in which the authors explore differences in network structure between internal and external researchers and the differences of research impact the two groups. However, the analysis is done on the aggregated-facility level and the beamlines only highlighted in the visualizations.

Material and methods

The European Synchrotron Radiation Facility (ESRF) is regarded as one of the leading synchrotron radiation facilities, and user facilities, in the world hosting around 4000 to 7000 annual visits for the period 2000–2018, with 1100–1900 experiment sessions per year (Cramer, 2017; ESRF Highlights, 2005, n.d.; ESRF Highlights, 2011, n.d.; ESRF Highlights, 2018, n.d.; Hallonsten, 2016b). Its central technological structure is the particle accelerator, or storage ring, which is responsible for accelerating electrons close to the speed of light. The radiation generated by bending the electrons in the storage ring is directed with and towards scientific instruments. It is in these bending magnets (BM) and insertion devices (ID) where the specialized instrumentation makes the beam usable to a large and wide range of scientists from different disciplines to perform a variety of specialized techniques to carry out their scientific work (ESRF, 2017; Hallonsten & Heinze, 2015; Hallonsten, 2016a; Söderström et al., 2022).

Facilities like the ESRF host not only their own research, experimentation and scientists but are also host to a large external multidisciplinary community which visits these scientific instruments—or beamlines—in the facility to do experiments as part of their ordinary scientific work. This distinction is a clear interest for such facilities, since part of their purpose is to support the external scientific community, and great effort is made into keeping statistics of their involvement in the facility. The facility is in a constant flux, introducing new instruments which may undergo both minor and/or major upgrading during their lifetime, and decommissioning them when they finalize their life cycle (Cramer, 2017; Hallonsten, 2009, 2016a, 2016b). The disciplinary range of the scientific instruments in the ESRF is wide, and it varies between the ca. 50 instruments currently hosted at the facility. Disciplines range from the likes of Chemistry, Applied Materials Science, Hard Condensed Matter Science and Structural Biology to other, although with lower representation, disciplines like Medicine, Earth Sciences, Environment and Cultural Heritage (ESRF Highlights, 2018, n.d., p. 172). Some of the insertion devices include the following:

• ID17: A tunable beamline for in-situ biomedical imaging, radiation biology and radiation therapy. Some of the disciplines that are covered by ID17 include medicine, cultural heritage, life sciences and materials/engineering. It is described as a medical beamline that is used to develop both imaging for diagnosis and irradiation therapy research (ESRF, ID[17], xxxx).

• ID19: A tunable hard X-ray beamline devoted to in-situ microtomography. An imaging technique used to create 3D models from 2D images. The techniques this beamline offers are said to be applied to a wide variety of disciplines like materials research, palaeontology, and industry. It is described as an instrument with high versatility capable of adapting to a wide range of experimental setups. (ESRF, ID19, xxxx).

• ID23-1: A tunable automated beamline dedicated to macromolecular crystallography experiments. A key technique in drug development. It is one of two independent end-stations operating concurrently: ID23-1, which is fully tunable, and ID23-2 which has a fixed energy level. Both are automated. Life sciences is the only discipline added in the beamline description by the ESRF (ID23-1, xxxx).

The Joint ILL-ESRF Library was used to collect bibliometric data on the publications using the ESRF, which included the beamline name and a binary variable that indicated whether the publication had at least one author employed at the ESRF. The full library dataset contains around 30,000 records for the period 1994 to 2018, consisting of journal articles, book reports, technical reports, and PhD theses. Since the interest lies in the differences in the co-authorship structure between beamlines, only publications that used one beamline are considered. The library data is matched with Web of Science via DOIs to make use of the highly structured data provided by the service and filtered to only include journal articles, which reduced the number of observations to around 15,000 with 57 different beamlines.

Beamline coverage, which was highly unbalanced, also needed to be treated in the dataset. Newer beamlines, older beamlines that went out of commission, and beamlines that received upgrades have lower coverage or are difficult to identify in the database due to their unique coding. For instance, the use of suffixes after a beamline name is common in the database, which complicates the classification of the beamlines for analysis. To achieve some balance between transparency, representation, and coverage, a selection of beamlines that accounted for the top 85% of the data was considered for the period 2000–2018. These constraints resulted in a dataset of 31 beamlines that account for 8323 publications.

The merged databases from the ESRF-ILL Join Libraries and Web of Science provide the analysis with the following relevant data for the current study:

• 1. From the ESRF-ILL Joint Libraries

  • a. Beamline name. For example, ID14, ID19, BM1, etc.

  • b. ESRF author(s): Boolean variable: TRUE if any author of the research output is affiliated to the ESRF, FALSE if no author is affiliated to the ESRF.

• 2. From Web of Science

  • a. Author names (AU)

  • b. Digital Object Identifier (DI)

From this original data in long format, grouping based on the beamline name was done to calculate the bibliometric and network structure data for every year. Finally, to calculate the network structure in data form, global descriptive statistics for each i = network and t = year are calculated. The construction and calculation of the network statistics is partly done with the Python package networkx (Hagberg et al., 2008).

This resulted in the following list of variables, used for the analysis:

• 1. Bibliometric data

  • a. Total number of publications by beamline.

  • b. Average team size by beamline: Total authors / total publications

  • c. Percentage share of external use by beamline

• 2. Network structure data

  • a. Number of nodes or vertices.

  • b. Number of edges or links.

  • c. Average degree: number of direct connections by node, as an average for the network.

  • d. Number of components: number of connected subgraphs in network.

  • e. Giant component size: share of nodes in largest component against total nodes.

The data is divided between a main sample (S1) and a subsample (S2). S1 includes the bibliometric and network structure data for all the scientific instruments or beamlines, where collection of descriptive statistics, time series visualisations and correlation analysis aid in finding how the network structures form and change over time, as well as finding relationships between the network structure and the external use of the instruments. The subsample (S2) includes a selection of three instruments and deepens the analysis with some qualitative descriptions. This subsample (S2) includes beamlines ID17, ID19 and ID23-1. Being a subsample of S1, instruments in subsample S2 are also present in S1.

The motivation for S2 was based on a preliminary analysis and the theoretical framework. First, beamlines with a high number of publications where selected (ID19 and ID23-1), but a comparison with a beamline with a lower number of publications (ID17) but with long tenure was also desired. Furthermore, ID23-1 with a high number of publications but relatively younger age, would be an interesting contrast with the longer running/high-output ID19 and the opposite longer running/low-output of ID17. Finally, the share of external users was also a motivating factor. ID23-1 shows a very high and constant share of use by the external scientific community, ID19 showing a lower but increasing share, and ID17 a constantly low share of external use.

Theoretical framework and hypothesis

The focus on instruments reflects de Solla Price’s (1984) instrumentalities, which inaugurated the modern focus on instruments in science studies. Together with the highly heterogeneous collection of scientific instruments in the ESRF motivates the framework for this study which narrows the focus to the collaboration network to the scientific instrument. Scientists positioned in central parts of the instrument collaboration network should be contain knowledge and know-how from the different teams of scientists in their proximity (Burt, 2004; Collins, 2012; Granovetter, 1973; Polanyi, 1962) and to the specific technological characteristics and related practices (Shrum et al., 2007). Knowledge of the scientific instrument and practices make internal scientists and staff an attractive collaboration prospect for incoming, external scientific teams. Moreso if the incoming team contains less experience in the surrounding the instrument, including the social environment, practices, and technology.

Thus, the instrument collaboration network focuses the role of scientific instruments in formalized collaboration and whether the knowledge and expertise of scientists in central positions in these networks are used by external teams with less experience with the instruments. With this framework in mind, the study explores the differences in network structures between instruments with different levels of external use.

Several hypotheses, or assumptions can be made based on this framework: The most connected node in an instrument network, and the outgoing connections from that node, reflects either practical knowledge of the beamline, some organizational leverage, and/or previous experience using the beamline. In contrast, a network that shows no central nodes or hubs and/or shows a highly fragmented network structure, with many unconnected components reflects a network in which little to no internal knowledge is needed, as no central connections exists between the diverse teams.

Results

Tables 1, 2, 3 show the beamline selection (S2) for further analysis including bibliometric and network structure data for the years 2005, 2010 and 2015. Figures 1, 2, 3 within Tables 1, 2, 3 show their respective network. The network visualizations and statistics are generated via the python library Networkx (Hagberg et al., 2008). Furthermore, the visualizations have fixed node and edge line sizes and use the spring layout algorithm available in the library.

Table 1 Instrument collaboration network statistics, beamline ID19

Table 2 Instrument collaboration network statistics, beamline ID23-1

Fig. 1

ID19 Microtomography Beamline

Fig. 2

ID23-1Gemini—Macromolecular Crystallography MAD

Fig. 3

ID17 Biomedical Beamline

Table 1 shows the network structure of beamline ID19 for the years 2005, 2010 and 2015. Visually, Fig. 1 shows a large giant component in the first 2 years of the comparison, while the last year that component is broken into three large components. It is possible to see central nodes that hold the three main large components together. The bibliometric data show an overall increase in publications, authors, and team sizes, it also shows an overall increase in the share of research done by external scientists. The network structure statistics show a similar growth in terms of size (nodes), connectivity and the number of individual components. However, the share of the largest component shows a relative decrease over the period. The ratio of components to publications increases over the same period.

Table 2 shows an example of a highly fragmented network in ID23-1 in the three comparison years, with no visually obvious hubs or centres of highly interconnected teams in Fig. 2. The bibliometric data indicate a newer instrument in terms of publications. However, it also shows a quick catching-up to ID19 in terms of size and output. Publications done by solely external scientists seem to dominate the instrument use since launch. The network statistics show high growth between the first and second comparison year, with increasingly higher connectivity and individual components. The share of the largest component and the ratio of components to publications show relative stability and reflecting the visualization of a highly fragmented structure.

Table 3 shows ID17, a highly connected instrument network over the selected period. Visually, Fig. 3 shows the networks as highly connected with minor variation over the period, other than an increase in size. The bibliometric data indicate a small instrument in terms of publications, although with comparable team sizes to the previous beamlines. Publications done by external scientists are low over the period, although they increased threefold in 2010. The network statistics show a small network with high connectivity. The small number of components, high share of the giant component and low ratio to publications show a dense and highly interconnected network, reflecting the visualization.

Table 3 Instrument collaboration network statistics: Beamline ID17

The beamlines in S2 are presented in Figs. 4 and 5 for the period 2000–2018 and compared to the mean of the full sample (S1) of the 31 beamlines, shown by the solid line, and the standard deviation shown by the shaded area in the graph¹. Figure 4 shows the bibliometric data, while Fig. 5 shows the network structure data².

Fig. 4

Source Author’s elaboration with data from ESRF-ILL Joint Library and Web of Science

Time series of bibliometric data. S2 comparison with S1 mean and standard deviation.

Fig. 5

Source Author’s elaboration with data from ESRF-ILL Joint Library and Web of Science

Time series of network data. S2 comparison with S1 mean and standard deviation.

Figure 4a–c shows the time series for the bibliometric data. Specifically, it shows the number of (a) publications, (b) average team size and (c) the share of external authors by year. The number publications have been increasing for the sample, as seen by the mean, with a wide variation, represented by the standard deviation. The three-beamline focus show an example of how different this growth rate is. For example, ID19 and ID23-1 show a high level of growth in terms of publications. ID17 in contrast, shows modest growth in terms of publications. Beamlines ID19 and ID23-1 show a rapid increase well above the standard deviation of the sample.

The average team size is steadily increasing over the period for the sample of 31 beamlines, ranging from around six person teams to averaging just under 8 person teams. There is variation between beamlines as seen by the three-beamline highlight. ID19 shows a similar starting point as the sample. However, it quickly diverges and retains a stable 5–6-person team size over its period, although with a slight positive trend over the period. ID23-1 shows team size and growth similar to the sample and mostly staying within the bounds of the standard deviation. Its average team size increases from around 6 to around 9. Lastly, ID17 shows a high team size at the beginning of the period, surpassing even the standard deviation of the original sample levelling off over time towards the mean.

In terms of the share of external use, the mean of the sample seems to be rangebound between 50 to around 75% of external use. However, the individual instruments have a wide range of internal/external use as shown by the standard deviation or shaded area. ID17 shows a constant low level of external users, below the lower end of the standard deviation of the sample. ID19, although also starting well below the rest of the sample shows a considerable growth in the share of external users over the period. ID23-1 on the other hand, shows a high level of external use from its first year in 2005.

Figure 5a–e shows the network statistics including (a) the number of nodes, (b) edges, (c) the average degree, (d) number of components, and (e) the size of the largest component as a percentage of all nodes. The network size of the sample has been increasing over the period from networks of about 50 nodes to over 100 nodes per year on average, mirroring the number of publications to an extent. However, ID17 shows smaller network sizes than the sample mean and beamlines ID19 and ID23-1, with bigger sizes, show growth from the sample mean of 50 to network sizes of about 350 nodes and 400 nodes. The latter surpassing 650 nodes in 2016. The number of edges has increased in a similar magnitude as publications for the sample mean. However, one key distinction in the explosive growth of ID23-1 relative to the rest of the sample. Somewhat related to nodes and edges, the average degree or connectivity of the sample steadily increases over the period form an average connectivity of around 5.5 to 8 average connections per node and seems to mirror the average team size in sample mean, standard deviation, and the subsample of beamlines ID17, ID23-1 and ID19.³

The number of components for the sample mean has increased from just under five components in the year 2000 to around 10 for the end of the period. The results shown by the standard deviation and the subsample show a wide variation between beamlines. For instance, ID17 show a consistently low number of individual components over the period, with single digit components for most of the period. ID19 is closer to the sample mean at the beginning of the period from around five components, but then diverge around 2013 to about 30 components at the end of the period. ID23-1 shows an opposite example to ID19, where the initial value of 10 components is quickly overcome coming to a more consistent 40 components per year to the end of the period.

The size of the giant component, as a share of total network size, has decreased from around 75 to 50% for the sample mean. However, there is also wide variation between the sample, and the subsample results suggest different initial positions and growth. ID23-1 shows a small giant component size, from 20 to under 10% for the rest of the period. ID19 and ID17 show an overall decrease of the size of the largest component over time closer to the sample mean. Overall, the giant component of ID19 shows to be smaller than the giant component for ID17 for the period.

The correlation matrix in Table 4 is calculated on the mean values by beamline⁴ to put the beamline differences into focus. The complete sample of beamlines (S1) used for the calculation of the Pearson correlation, which identifies the presence of a linear relationship between variables.

Table 4 Correlation matrix

The share of external users is highly correlated with the share of the largest component (− 0.83***), the number of components (0.63***); somewhat correlated with the average degree (− 0.43***), publications (0.4***) and the size of the network (0.49***), all showing a high level of statistical significance. The association between external use and network structure is visualized in Fig. 6a–c below, which include (a) the share of the largest component, (b) the number of components and (c) the average degree.

Fig. 6

Source Author’s elaboration with data from ESRF-ILL Joint Library and Web of Science

Relationship between network structure and external use.

There is a high correlation (0.89***) between average team size and average connectivity as expected from the time series plots. A low to null correlation between team size and largest component share (0.22), number of components (0.07) and network size (0.13). Except for the high correlation with average degree (0.89***).

There is a high correlation between the number of publications and network size (0.92***), number of edges (0.81***) and the number of components (0.82**), medium to high correlation with the giant component (− 0.59***) and external use (0.4**) and null correlation with average team size (− 0.15) and degree (− 0.14).

Overall, the results show a significant relationship between the level of external use and the structure of the network at various levels of analysis. The first section of results included simple visualization and descriptive statistics that showed this relationship. For example, beamline ID23-1, with an elevated level of external use for all its lifecycle, is associated with a highly fragmented network structure over the same period as shown in Fig. 2. The opposite is also apparent in Fig. 3 for beamline ID17, where the highly connected network reflects a low level of external research. The final example, shown by ID19 shows a growth of both the external share of research and the number of components and reduction of the share of the largest component. The second section of results analyses the change of the three beamlines (S2) and compares it with the main (S1) sample mean and standard deviation to analyse individual and general trends of the bibliometric and network structure data, finding how the different instruments grow and change over time. Finally, the third results section attempts to find relationships between the instrument data in aggregated form and finds strong correlation between the level of external use and the structure of the network. While this loses the time domain in the analysis, it avoids spurious correlations between potentially time-dependent or autocorrelated variables.

Discussion

The theoretical perspectives (Burt, 2004; Granovetter, 1973) that informed the focus on the instrument collaboration network imply that central nodes are important to connect different groups with related, but distinct and valuable, knowledge (Collins, 2012; Polanyi, 1962). A priori expectations would have anticipated a positive relationship between share of external use and tighter network structures, showing that the external scientific community would collaborate formally, in the form of co-authorship, with the internal scientists or staff to access the knowledge required for use. However, this was not met by the results of this study. Instead, the results show the opposite relationship between the level of external use and network structure, showing more interconnected networks with low and midlevel external use. Whenever the external community dominate the use of the instrument, the teams in the networks were mostly disconnected from each other. As Shrum et al. (2007) suggested, differences in technology and practice might explain the differences in structure, which seem to be observed in the results.

While a small selection, the subsample (S2) of beamlines ID23-1, ID19, and ID17 might motivate further inquiry into the role technology plays in collaboration. The openness of an instrument, or how generic it is, could play a role in network structure, just as it plays a role on the range of disciplines it can cater to (Söderström et al., 2022). A beamline like ID23-1, with a high level of external use and highly fragmented network structure for most of its lifecycle, is presented as a beamline that supports automated processes that would potentially make it easier for external scientists, not familiar with the technology to make use of it. A beamline ID17, described as performing in-situ experimentation seems to be designed to host a more homogeneous community in the medical sciences, reflected in their constant level of connected components and high share of concentration of nodes in the largest component. ID19 is a beamline whose structure over time seems to change, opposed to the previous examples. The beamline is described as an instrument with a wide variety of uses with high level of versatility. Suggesting perhaps adoption by the wider scientific community over time.

The results also show that the instruments in the facility are different form each other, to the extent that aggregating an overall analysis of a facility without addressing their heterogeneity could bias results. The findings support previous research that highlights a view of the instrument as the unit of analysis and that disciplines vary between and within the instruments, likely another decisive factor into the collaboration structure (Hallonsten & Heinze, 2015; Mody, 2011; Söderström et al., 2022), the importance of the time component in the analysis of networks (Holme & Saramäki, 2012; Leskovec et al., 2007) and its consequences on the structure of collaboration between scientists (Glanzel, 2001; Melin & Persson, 1996; Shrum et al., 2007).

The results are in line with the related research. For one, a study of the Swiss Light Source, the number of isolated communities and the giant component in terms of percentage share also show different in the network structure between the internal and external scientist networks. While the focus of said study is on differences of impact and network structure of the aggregated internal and external users, it shows visualizations in the study which attempt to differentiate various beamlines (Silva et al., 2019). Another study that compares internal and external research staff by di Bella et al. (2021), also finds structural differences between the respective networks.

The findings show some contrast with previous research of networks with disciplinary or institutional groupings, which usually find a giant component that occupies around 80 to 90% of all authors. Most of the scientific community in the studies are highly connected, with an average of six degrees of separation between a pair of scientists in the network (Newman, 2001, 2004a, 2004b). Some findings that look at dynamic networks find average degree increasing over time, with node separation decreasing, and growth determined by preferential attachment. (Barabasi et al., 2002). In contrast, this study finds that while average degree of the instruments at the facility is indeed 6 at the beginning of the period, it increases steadily over time with a wide variation between instruments. The analysis of preferential attachment is not central to this study. While a similar mechanism exists in instruments with internal scientists and staff, this phenomenon seems absent when external researchers dominate the use of the beamline. While this result is expected in hindsight, it highlights the importance of understanding what tools or technology the community being studied uses. Furthermore, while Bettencourt et al. (2009) find a topological transition from small, disconnected graphs to larger networks where a giant component of collaboration appears, this effect is only present in some instruments, mainly the instruments with persistent, internal staff that connect different teams in the publication outputs, like ID17 or ID19.

Co-authorship can be influenced by factors such as the norms and expectations of different fields, the motivations of individual researchers, and the incentives and rewards associated with publishing research. Therefore, the use of co-authorship as a measure of collaboration may not accurately reflect the full range of ways that scientists interact and share knowledge. However, within the context of this study, it might shed light into how willing external scientists are to include internal scientists as co-author in terms of their perceived amount of work for their own study. It would be safe to assume that every instrument or beamline requires a great amount of work by internal scientists and staff to build, operate and maintain. However, in some cases, like where scientists can mail in samples and operate the machinery on their own or even remotely, they might feel less inclined to include them as co-authors, when compared to instruments where more face-to-face collaboration and planning might be needed. Whether these differences in the formalization of collaboration are fair or expected is beyond the scope of this study, but it adds an interesting question surrounding the practices and rewards of collaboration among scientists.

Conclusions and future work

This study analyses the collaboration network of scientists on the instrument level of a synchrotron radiation facility with the aim of understanding how constellations of scientists group and align within and between instruments. It also illustrates how these structures change over time. The results show significant differences in the structure of the networks between instruments, with an important factor being the level of external use, an important characteristic of facilities like the European Synchrotron Radiation Facility (ESRF), and other so-called user facilities.

The results of the study appear to show a significant relationship between the level of external use of scientific instruments and the structure of the collaboration networks associated with these instruments. This relationship is demonstrated through visualization and descriptive statistics, as well as more in-depth analysis of the change in network structure over time and the correlation between the level of external use and network structure in aggregated form. These findings suggest that there may be a strong link between the level of external access to scientific instruments and the collaboration networks that form around them, with instruments that have a higher level of external use being associated with different network structures compared to instruments with less external use.

While the quantitative analysis could not provide reasons for this, a brief discussion based on the instrument descriptions and the theoretical framework suggests considering the technological and practice related characteristics of the beamlines into account, providing an avenue for further inquiry. Perhaps the technological aspects of the instruments, together with disciplinary differences between the uses of the instruments play an important role in how scientists collaborate and also formalized that collaboration in the form of co-authorship.

Considering the success of the ESRF as a user facility and its place as one of the premier synchrotron facilities, perhaps these results could shed light into other similar facilities or projects where external users make up a high share of the userbase. This could have important implications for the way that scientific instruments are used and the level of collaboration between different teams and disciplines in similar facilities or projects.

Appendix

See Table5.

Table 5 Bibliometric and network structure data, beamline average