Why do some retracted articles continue to get cited?

Abstract Retracted publications can still receive a substantial number of citations after the retraction. Little is known about the causes for this phenomenon and the nature of epistemic risk or harm in these retraction cases. Using this phenomenon also as an example for the broader question of how scholarly communities deal with uncertainty in the reception of publications, this case study aims to assess the epistemic contributions of retracted publications with continuous and decreasing citation impact and to relate these to the epistemic environments of the retracted papers and to reception patterns. Several parsing and natural language processing approaches are used, complemented, and validated by qualitative close reading. Specifically, (i) dissent and support are identified in citing and in co-cited publications; (ii) the concept terms of retracted publications are expanded by word embeddings and MeSH terms and traced in citation contexts; and finally (iii) rhetorical functions in citation contexts are identified based on keyword extraction. Empirical support and unresolved disputes are found almost exclusively in cases with continuous citations. Authors emphasize specific informational values in some cases with citations continuing after the retraction, while methodological and more general levels of claims prevail in others. Citations can be meaningfully examined by considering the weighing process between epistemic risks and informational value; persistent citation impact thus doesn’t necessarily indicate the perpetuation of epistemic harm.