Reference behavior in the full text of scientific articles: A large-scale analysis

In this paper, we report distributions of in-text citations in full text articles from two large databases - the PubMed Central Open Access Subset and Elsevier journals - as functions of time and textual progression. Using over five million articles from these two databases, we show that the numbers of in-text citations (i.e., mentions) per reference have remained nearly constant over time, and that the number of mentions per reference decreases with reference age. The data also show that mentions are concentrated at the beginning and near the end of articles, but that the distribution changes with the number of mentions per reference. References ages peak on average at around the 30% point in an article, and decrease thereafter. Citation counts for references also peak at around the 30% point, suggesting that methods papers are more highly cited than other types of papers. Citation counts are also a function of mentions - those references mentioned only once are much more highly cited than those mentioned many times, which correlates with the notion of perfunctory references being mentioned less often than those references that are more central to a citing article.