The Most Cited Papers Are Not the Best Starting Points: What the Reception of a Paper's Citing Papers Tells an Author
Year: 2026
Type: preprint
Abstract: Manuscript, state of 13 November 2026. The file is restricted; this record documents the state of the text at this date. Abstract. Which papers of an established field are good starting points for new work? Papers that build on the most cited papers are received hardly better than average. We test a rule that follows from Newman's hope that papers standing out against their contemporaries do so by their quality: rank papers by the mean cohort z-score of the papers that cite them. In two fields of arXiv, artificial intelligence and mathematics, 12.6% and 8.7% of the papers that newly cite a recent paper become strongly received. For the top 2% by our rule the shares are 20.6% and 16.5%; by Newman's z-score 13.8% and 10.9%; by time-rescaled PageRank 12.6% and 8.8%. Against controls of the same category, citation count and age the ratio is 1.47 and 1.48; it is 1.34 and 1.30 when the citing authors have the same record of earlier reception, and 1.12 in artificial intelligence within the same author. The evidence is observational, and the rule does not predict how often the paper itself is cited. All 70 predictions were sealed and their checksums deposited before each measurement; thirteen failed. The rule specified in advance as primary held but adds nothing detectable to the simpler rule. The checksums of the sealed analysis plans and of the programs are deposited separately: doi:10.5281/zenodo.24138400 (all versions).
Author: von Hilgers, Philipp
Language: English
Subfield: Library and Information Sciences
Field: Social Sciences
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7226029191
