Fifty Years of Data Reveal How the World’s Academic Migration Network Went Global

Academic migration has long been described as a one-way brain drain from poorer countries to richer ones. A new network analysis of nearly five decades of scholarly movement suggests that picture is wrong at the global scale: the world’s academic migration network is overwhelmingly circulatory, with countries that exchange researchers at all almost always doing so in both directions.

The study, published in Applied Network Science, reconstructs the worldwide academic migration network from 1974 to 2023 using affiliation changes recorded in almost 50 million Scopus articles. Taylan Yenilmez, the author, tracked the countries associated with each author’s publications over time, treating a change in country of affiliation between consecutive observed years as an inferred migration dated to the year of arrival. The dataset covers 49,335,767 articles by 32,252,350 individuals, of whom 2,317,361 had more than one country of affiliation during the study period.

Rather than analyzing year-by-year flows, the study aggregates annual migration matrices into five decade-long snapshots covering 1974–1983, 1984–1993, 1994–2003, 2004–2013, and 2014–2023. The earliest years, 1970 to 1973, were excluded because migration is inferred from affiliation changes over time, and those years lack a prior window to establish a baseline. Countries were treated consistently across political upheavals: Soviet-era flows were unified under the USSR, Yugoslavia and Czechoslovakia were handled as single entities until their dissolutions fell at a decade boundary, Germany’s pre-unification states were merged throughout, and Hong Kong was combined with China.

The headline structural finding is densification. At the zero threshold, where every observed flow counts as a link, network density almost triples over the five decades, from 12 percent in the first decade to 32 percent in the last. The same monotone rise holds at the 0.001 percent threshold, where density grows from 8 to 11 percent, and at the 0.01 percent threshold, where it moves from 2.9 to 3.3 percent. The direction reverses only at very high thresholds: at 0.1 percent density falls from 0.67 to 0.51 percent, and at 1 percent from 0.07 to 0.02 percent. That reversal is not a contradiction, the author argues, but a consequence of the same process, because as new hubs emerge and flows spread across many corridors, fewer individual links cross the very high share thresholds even while the network as a whole grows denser.

A natural concern is that the apparent growth in connectivity simply reflects the exponential increase in the number of published articles, which could mechanically inflate recorded affiliation changes. To address this, the decade networks were rebuilt from subsamples in which the number of authors was held equal to the active author count of the 1974–1983 decade, and all metrics were recomputed. The results were similar, indicating that the reported patterns are not an artifact of the growing author base. The analysis also uses shares of global migration rather than raw counts as network weights, so results focus on changes in network structure independent of absolute growth in migration numbers.

Reciprocity emerges as the most robust property of the system. It never falls below 76 percent at the zero threshold and exceeds 90 percent at the 0.01 and 0.1 percent thresholds in the early decades. Because high reciprocity holds across the full fifty-year span and across every threshold choice, the author concludes it is a property of the academic migration process itself rather than an artifact of how the network was constructed. Consistent with this, the most central countries are simultaneously major senders and major receivers: the United States is the top destination and the top sender in every decade, the United Kingdom ranks second on both measures throughout, and Canada, Germany, France, and Australia appear in the top ten on both incoming and outgoing flows in every decade.

The United States remains the network’s singular center for the entire period. Its weighted in- and out-degree centrality declines from roughly 30 percent in the 1970s to roughly 20 percent in the 2020s, meaning one in three to five academic migrations in the data involves the United States. Its strongest ties have consistently been with large economies such as the UK, Canada, Germany, France, and Japan, with China emerging as a key partner in the last two decades. Yet the decline in the American share is part of a broader fall in inequality: the Herfindahl-Hirschman Index of weighted in- and out-degree centralities has steadily declined across all five decades, as other countries, particularly in Asia, have become increasingly significant nodes and new local hubs have emerged.

The most detailed story within this core concerns the rise of Asia. China is absent from the top 20 in 1974–1983, enters the incoming-flow ranking at rank 15 in 1984–1993, climbs to rank 11, then 6, and reaches rank 3 by 2014–2023, with the outgoing ranking tracking the same trajectory to rank 4. South Korea enters the top 20 on incoming flows in 1994–2003 and holds rank 15 on both measures in the final decade, while India is present in the top 20 throughout and moves to rank 12 on incoming and rank 7 on outgoing flows in the most recent decade. Crucially, no separate Asian cluster forms at any point. Cluster analysis at the 0.1 percent threshold shows a persistent two-part split between a US-led cluster covering most of the world and a European cluster anchored by Germany and France, and from the 1990s onward China, South Korea, Taiwan, and Singapore join the US-led cluster rather than forming their own. By 2014–2023, China sits at the core of that cluster as a top node alongside the United States, and ranks second behind it in HITS authority score.

The network’s average shortest path length is around 2 at almost all thresholds, indicating that most countries connect to each other through a few central hubs rather than directly. The author notes that an average path length of 2 could in principle also describe a uniformly dense network, but the moderate densities of 12 to 32 percent rule that out: at those densities a typical pair is not directly linked, so short paths must be mediated by a small set of high-degree countries. The study also applies HITS authority and hub scores, which in this context capture brain-gain and brain-drain roles respectively, though recent work suggests such eigencentrality measures lose discriminating power in highly reciprocal networks, and indeed the HITS rankings here track the weighted degree rankings almost exactly.

Important limitations accompany these findings. A change in academic affiliation does not correspond directly to physical migration, since sabbaticals, short visits, and joint appointments can all produce affiliation changes without relocation, and the study’s adjacent-year rule mitigates but cannot eliminate this problem. The restriction to English-language articles means researchers who publish exclusively in other languages are invisible, underrepresenting mobility in countries such as China and Brazil; moreover, because the share of English-language output from non-Western countries rose during the study period, some of the apparent rise of these countries may reflect increased English-language publishing rather than genuine mobility growth, a bias the constant-author-count robustness check only partially addresses. With those caveats in place, the study frames its findings as the structural signature of globalization: a network that has densified, become less concentrated, and remained circulatory, with the open question being whether these trends will persist or reverse as a geopolitical turn toward deglobalization unfolds.

Subject of Research: Reconstruction of the global academic migration network from affiliation changes in Scopus articles, 1974 to 2023

Article Title: Fifty years of the world academic migration network

Article References: Yenilmez, T. (2026). Fifty years of the world academic migration network. Applied Network Science, 11(1), Article 65. https://doi.org/10.1007/s41109-026-00840-0

Image Credits: AI Generated

DOI: 10.1007/s41109-026-00840-0

Keywords: academic migration, network analysis, Scopus, globalization, brain drain, reciprocity, China, United States, bibliometrics, centrality, higher education, scientific mobility