In recent years, bibliometric databases have substantially improved the consistency and quality of the metadata extracted from publications, particularly the author-affiliations linkages from scientific publications. Furthermore, author-name disambiguation algorithms have been implemented for most large bibliometric databases, such as Web of Science, Scopus, and Dimensions. Most of these algorithms benefit from open systems for the unique identification of scholars like ORCID. These developments have led to new scientometric approaches to track aggregated mobility patterns between countries and regions.

However, when it comes to the mobility of researchers between institutions, evidence is scarce due to the lack of harmonized affiliation data at the micro level. This situation is changing due to the implementation of advanced approaches for affiliation harmonization like the approach used for the Leiden Ranking, or, more recently, the Global Research Identifier Database (GRID), that currently covers more than 98,000 research institutions worldwide. The availability of these harmonized registries enables the identification of affiliation changes in the careers of scholars. In this blog post, we illustrate how author-affiliation information can be used to develop mobility indicators at the institutional level and explore the many possibilities this offers for the study of scientific mobility.

Measuring the institutional mobility of scholars

There are many different forms of mobility, and the interpretation of such forms may vary from case to case. To keep things simple, we focus only on the distinction of institutionally mobile vs. non-mobile. Specifically, we consider that a researcher is institutionally mobile when she is affiliated to more than one institution in a given timeframe. Based on this broad definition, we categorize researchers into two different mutually exclusive groups:

• Non-mobile, researchers who have been only affiliated to one institution during the period of analysis.
• Mobile, researchers who have been affiliated to more than one institution in the period of analysis.

The dataset: identifying researchers and their institutions

We use the Dimensions data dump updated until June 2019 that is available at CWTS, selecting only publications from the period 2015-2018. This time period is the same as the one used in the most recent release of the Leiden Ranking.

Mobility research using bibliometric data relies on the connections between authors and publications. Dimensions data allow the authors to be directly linked to their publication level affiliations. In order to link publications with individual scholars, Dimensions implements an author-disambiguation algorithm, allowing to know which publications were authored by which author in their database. This algorithm has a higher level of precision (i.e. How many identified publications truly belong to the given researcher?) than recall (Are all researchers’ publications correctly identified?). While higher precision is a desired property for an author-disambiguation algorithm, the lower recall may lead to underestimation of the number of mobile researchers, since some of their publications (and affiliations) may not be identified.

In order to track scholars’ movements at the institutional level, we combine the affiliation information of those articles with the unique institutional affiliation names available in the GRID. Given the large scope of organizations included in GRID, and in order to work with a homogeneous set of institutions, we crossmatched 1,176 distinct Leiden Ranking institutions with GRID identifiers. We found a total of 1,169 organizations that had a GRID identifier and appeared in the latest release of the Leiden Ranking (2020). Therefore, we adopt the methodology of the Leiden Ranking for the identification of institutions. This limits our analysis to a set of well-established universities.

Global institutional mobility

Figure 1 shows the propensity to mobility within universities included in the Leiden Ranking. The results indicate that North American, Australian and Northern European universities exhibit a relatively high level of institutional mobility. Countries in Southern and Eastern Europe show lower levels of institutional mobility, something that is also observed in South America, Africa, and Asia, including Japan and South Korea.

An important element to consider when studying mobility with bibliometric indicators is to which extent publications of researchers provide reliable signals of mobility. In order to test this for the results in Figure 1, we performed a test studying only those researchers with exactly 5 publications. This returned patterns very similar to those in Figure 1.

So far we analyzed only mobility events for Leiden Ranking universities. This approach allows tracking mobility among a relatively homogeneous set of institutions but does not capture the mobility events of researchers with institutions that are not covered by the Leiden Ranking. For example, we did not track mobility linkages of researchers with other local universities and with other types of research organizations in the government sector and elsewhere (e.g. thematic or umbrella research organizations like Inserm, CSIC, CNRS, CNR, Instituto de Salud Carlos III, national academies of sciences, etc).

In Figure 2, we study the mobility of researchers affiliated with universities covered by the Leiden Ranking, while also considering their mobility events with any other institution identified in GRID. This means that mobile researches are not only those that have been affiliated with more than one Leiden Ranking university, but also with any other organization identified by GRID. This includes, for instance, government research organizations, umbrella research organizations and their subsidiaries, as well as hospitals.

Overall, the general trends observed in Figure 2 remain the same as in Figure 1. However, this time we see a stronger propensity of mobility particularly in French and Italian universities that may be related to the important role of national research organizations such as CNRS (France) and CNR (Italy) in structuring and developing those countries' scientific occupations. Hence, one needs to keep the impact of these organizations in mind when analyzing differences in mobility patterns between individual countries, but the broad geographical patterns are robust to this bias.

Expanding mobility analytics

The devil is in the detail. While the previously outlined methodology is simple to calculate and understand, it may be challenging to interpret in the light of more refined and detailed mobility concepts, like migration, brain drain, etc., which have more relevance from a policy-motivated point of view. For example, the indicators above may confound cases in which a given researcher is marked as mobile because she simultaneously keeps multiple affiliations, or because she moved towards new opportunities in a different institution. The short timeframe of analysis may also be limiting. Policy-motivated research will require more directed and focused indicators.

That’s why we also want to offer a quick look at what we could see as a potential indicator of inbreeding. This indicator is motivated by the previous study "Where do universities recruit researchers from?" Assuming that researchers most likely publish their first paper(s) with affiliation to their Ph.D institution, this indicator aims at measuring the tendency of universities to keep their own Ph.D graduates affiliated to them over time. We first identify researchers who published with any given Leiden Ranking university in 2018 and collect their full publication history (as in Dimensions). We excluded academically younger researchers by removing all researchers with the first publication after 2012, thus retaining only those researchers who started to publish in or before 2012. Then we classified researchers into two types:

• Insiders: Researchers whose first affiliation and their affiliation in 2018 remain the same.
• Outsiders: Researchers whose first affiliation and their affiliation in 2018 are different (i.e. in 2018 they are in a different institution than when they started to publish).

Interestingly, Figure 3 exhibits a very similar pattern to those shown previously. While U.S. and Western European universities have (in 2018) researchers who more frequently started to publish in other universities, this is much less common in Eastern Europe or Southern Europe. For example, only 1 out of 4 researchers at Indiana University Bloomington started their publishing career at the same university. This figure is 3 out of 4 at Sapienza University in Rome or the University of Warsaw. Readers can study the data in more detail in an interactive dashboard depicting differences between institutions and fields of research.

The way forward

The preliminary analysis presented in this blog post supports the possibility of monitoring and studying mobility patterns at the institutional level using bibliometrics. Many perspectives and alternative indicators can be formulated to fully understand global mobility patterns. Issues related to the economic, social, reputational, linguistic, geographical, generational, systemic, or political aspects of scientific mobility (e.g., youth and size of the different scientific systems, travel bans, war conflicts, or the effects of crises like the current COVID-19 pandemic, etc.) are all elements that need to be considered in future scientific mobility studies. Moreover, more advanced conceptual definitions of mobility (e.g. migrants vs. multiple affiliations, inbreeding, or brain drain/brain gain processes) are also necessary for more advanced policy-relevant studies of the different dynamics related to scientific mobility.

Finally, more technical issues such as the coverage, completeness, and accuracy of the Dimensions database and of the GRID identifier, or more conceptual aspects such as the importance of scientific mobility in creating new thematic flows across institutions will deserve our attention in the near future.