Observational Author Contributions in arXiv Papers

This dataset provides author-level contribution information for a large corpus of multi-authored arXiv papers submitted between 1991 and 2023. Contributions are inferred using behavioral writing signatures based on LaTeX macro usage. Each record includes section-level document structure and estimated contributions for each author. The dataset was constructed from 2.8 million LaTeX source files collected from arXiv.org, encompassing over 2 million unique authors across major STEM disciplines. This resource supports large-scale studies of scientific collaboration, authorship patterns, and the division of labor in scholarly writing.