Persistent anomalies in the Stanford/Ioannidis top 2% scientist’s database: empirical impact of publication-type filtering, temporal coverage correction, and retraction-rate thresholds

Abstract The Stanford/Ioannidis database of standardized citation indicators, now in its eighth version (August 2025), ranks over 230,000 scientists in the top 2% by composite citation score. While this resource represents a landmark contribution to research evaluation, our exploratory analysis of the career-long dataset reveals several persistent anomalies that require attention. We identify five categories of concern: (1) the inclusion of science journalists as top-ranked single-authored “scientists,” mixing news reporting with peer-reviewed research; (2) historical coverage artefacts in scopus data, where historical scientists (some deceased before 1960) exhibit impossible single-authored/total-publication ratios exceeding 5000%; (3) the absence of a retraction-based exclusion threshold, permitting researchers with retraction rates above 34% and retraction-derived citations exceeding 75% of their total to remain listed; and (4) ranking discontinuities in the lower portion of the list, where scientists at the tail end of the top 2% occupy actual rank positions around 1,300,000, with career metrics (mean h-index of 10) that appear inconsistent with top-percentile designation, (5) the presence of hyper-prolific authors whose output consists predominantly of non-research publications such as editorials and letters, yielding approximately 1.3 citations per paper despite listing nearly 5000 publications. Going beyond identifying these anomalies, we implemented each proposed correction algorithmically on the full V8 dataset and report the empirical consequences: rank stability was assessed using Spearman and Kendall rank correlations, retention rates were computed for the top 1000, 10,000, and 100,000 scientists under each filter and their combination, and a sensitivity analysis was conducted for retraction-rate thresholds ranging from 1 to 20%. Even under combined filtering, 99.30% of the top 1000, 98.92% of the top 10,000, and 97.47% of the top 100,000 scientists were retained, with perfect rank-order preservation (ρ = 1.000, τ = 1.000).