Identification of hyperauthorship in scientific publications: an approach based on outlier detection methods

Abstract

The phenomenon of hyperauthorship—characterized by excessive numbers of co-authors in scientific publications—has emerged as a significant concern in contemporary bibliometric analysis. While the growth of collaborative research is well-documented, systematic methods to identify what constitutes "excessive" authorship remain underdeveloped. This study presents a comprehensive approach to hyperauthorship identification based on outlier detection methods, analyzing over 52 million publications from Scopus (2003–2024) across 310 categories. A preliminary analysis of this data reveals substantial temporal and disciplinary variations in authorship patterns, with single authorship declining from one-third of publications in 2003 to significantly lower levels by 2024, while multi-authored works (>10 authors) increased dramatically. Then, 14 different outlier detection methods were implemented, including classical robust measures, skewness-corrected approaches, sequential methods, clustering techniques, and a parametric model based on the discrete power-law distribution (POW). This last method demonstrated superior performance, producing suitable thresholds in 100% of cases without indications of potential over- or under-identification of outliers. The systematic application of POW reveals hyperauthorship thresholds ranging from 3 authors in early 2000s Social Sciences to over 40 in recent Medicine and Physics publications, with an overall increase in hyperauthorship rates from 0.81% to 1.01% globally between 2003–2024. Medicine showed the most dramatic evolution, nearly doubling its hyperauthorship rate to 1.60% by 2024, while areas like Energy maintained consistently low rates (0.53%). These findings provide evidence-based reference points for editorial policies and research evaluation, demonstrating that hyperauthorship assessment requires discipline-specific and temporally-adjusted approaches rather than universal thresholds.