A global perspective on social stratification in science

Scientific inequality is a significant issue, influenced by factors such as institutional resources, gender, race, ethnicity, migration status, and socioeconomic background. While bibliometric measures like publications and citations are often used to assess success, they are typically examined in isolation. This paper addresses this gap by providing a comprehensive assessment of stratification across scientific fields using a multivariate analysis of large-scale bibliometric data (1996-2021). The study argues that the interrelationships between bibliometric indicators offer a more nuanced understanding of inequalities than single-variable analyses. A dataset with country-level measures of scientific stratification is made publicly available for future research.

Existing research indicates growth in scientific activity and geographical expansion, along with increased co-authorship and geographical mobility. However, this growth is accompanied by a concentration of success indicators among a small group of scholars. Data from Scopus (1996-2021) reveals that a substantial portion of scholars have limited publication records, suggesting that a few highly productive researchers drive overall productivity trends. Similarly, citation distributions are skewed, with a small share of publications receiving a disproportionate number of citations. These studies highlight the need for a multidimensional approach to understanding inequalities in science, considering the interplay between productivity, collaboration, mobility, and citations.

The study utilizes data from 28.5 million publications indexed in Scopus (1996-2021). Author disambiguation was crucial, leveraging Scopus identification numbers and the Research Organization Registry (ROR) API to link publications to authors and geocode affiliations. The analysis focuses on 8.2 million disambiguated authors. Twelve bibliometric indicators were selected to capture productivity, collaboration, mobility, and visibility. These indicators were computed at the author level, encompassing their entire career. To ensure comparability, most indicators were standardized by authors' academic age (years since first publication). The data was categorized into multiple classes to mitigate the impact of outliers, preserving distributional characteristics. Multiple Correspondence Analysis (MCA) was used to generate factorial axes summarizing the relationships between indicators. Hierarchical clustering (Ward method) followed by K-means algorithms identified six bibliometric classes (bottom, low, mid-low, mid-high, high, top). Network analysis of co-authorship networks, using the Constant Potts Model (CPM) and other community detection algorithms, identified collaboration communities and assessed the distribution of authors across bibliometric classes within these communities.

The study reveals a stratified structure in all six macro fields of science. A small top class (6-19% of authors) accounts for a large share of academic outputs in terms of productivity, impact, and collaboration. However, bottom classes contribute similarly or even more to mobility indicators (national and international moves, number of affiliated organizations). The first three MCA axes capture the main dimensions of bibliometric performance: "Academic age, number of organizations, and individual productivity"; "Total productivity, visibility, and collaborations"; and a third axis largely reflecting productivity and collaboration, with field-specific variations. The 20/80 rule, suggesting that the top 20% contribute 80% of the output, is not generally supported by the multivariate analysis; the top classes consistently contribute less than 80% to most indicators. The bottom classes contribute minimally to most indicators except for mobility measures, indicating that mobility is associated with both success and failure in academic performance. Network analysis shows that collaboration communities comprise authors of all ages and bibliometric classes, though stratification based on age and bibliometric classes show striking similarities.

The findings highlight the complex nature of scientific stratification, challenging simplistic interpretations of success based on single indicators. The multidimensional approach reveals a persistent hierarchical structure across scientific fields and age groups, with a disproportionate concentration of success among a relatively small elite group. The similar contribution of bottom and top classes to mobility measures indicates that mobility is a double-edged sword. While it can lead to success, it can also be associated with instability and failure. The persistent stratification and the disproportionate benefits of collaboration for the top class raise important questions about equity and access within science.

This study provides a comprehensive, globally-focused analysis of social stratification in science, revealing a persistent and stratified structure across fields and age groups. The findings challenge the 20/80 rule and highlight the complex interplay between various bibliometric indicators. Future research should investigate the underlying causes of this stratification, including resource allocation, mentorship, and the impact of national research policies. This work emphasizes the need for a more nuanced and equitable approach to assessing and promoting scientific success.

The study's descriptive nature limits causal inferences regarding the underlying mechanisms of scientific stratification. While the study includes a broad range of bibliometric variables, it does not account for factors such as the prestige of institutions, the type of research contracts, or the availability of research assistants. These limitations could influence the interpretation of the findings and the generalizability of the results.