TY - JOUR AU - SUN, Shiyue AU - LI, Jinli AU - DENG, Yan AU - ZHANG, Baohui AU - GUO, Haiqiang PY - 2026 TI - An Analysis of Data Paper Retractions and Their Implications JO - Science-Technology & Publication SN - 1005-0590 SP - 105 EP - 114 VL - 45 IS - 9 AB - As a publication format supporting open science, data papers have emerged as a formal publication type that systematically describes scientific datasets, enhancing their visibility, accessibility, reusability, and citability. Unlike traditional research articles, data papers provide value through datasets and their processing workflows, so any quality issue may propagate along reuse chains, creating persistent data contamination beyond the invalidation of a single conclusion. Retraction dynamics of data papers remain largely unexplored, with prior studies focusing on conventional articles. To address this, we retrieved all data papers indexed in Web of Science from 1900 to 2026, identifying four retracted cases. For each, we collected bibliographic metadata, retraction notices, and Retraction Watch records, and manually examined all citing articles to code citation location and intent, using a taxonomy that distinguishes substantive (data-oriented or method-oriented) from non-substantive background citations. Post-retraction citations were classified as informed or uninformed, with a one-year grace period to account for publication delays. Our analysis indicates a retraction rate of 0.066%, below the global average of 0.3%, but this low rate does not indicate superior integrity; rather, it may reflect the difficulty of detecting errors that can occur at any data stage and the passive, reuse-dependent discovery process. The four cases exhibit diverse causes: two involve traditional misconduct, while the other two stem from issues specific to data papers, namely loss of third-party permission and severe data contamination. Retraction delays range from one to four years, with some quality problems taking longer to surface. In terms of impact, the four papers received 132 citations, 89.4% pre-retraction; among 14 post-retraction citations, 92.9% were uninformed, suggesting limited awareness of retraction notices. Citations are concentrated in Materials and Methods (41.4%), followed by Introduction (28.4%), Discussion (17.9%), and Results (12.3%). Substantive citations account for 54.9%, and among substantive citations, method-oriented citations (67.4%) far outnumber data-oriented ones (32.6%), suggesting that the data papers in this sample are often used as sources of analytical workflows. In the records examined, monitoring systems such as Retraction Watch lacked a dedicated data-paper category. Publishers should embed automated retraction checks into submission, peer review, and production processes. Data repositories must link datasets to papers, display prominent warnings, and notify prior downloaders when retraction occurs. For limited errors, versioning and corrigenda should be preferred. When data are fundamentally unreliable or their use is legally non-compliant, retraction should include detailed guidance. We propose a full-lifecycle quality control framework covering data generation, cleaning, validation, publication, and updates, including mandatory quality reports, integrity-focused review, compliance commitments, and regulatory tracking. Journals should assess studies relying on retracted data and issue corrigenda or retractions as needed. Given the small sample and limitations of Web of Science coverage, our findings are exploratory and need further validation. Nevertheless, data paper retraction governance requires collaboration among stakeholders to safeguard data reuse and support open science through clear standards. UR - https://doi.org/10.16510/j.cnki.kjycb.20260929.004 DO - 10.16510/j.cnki.kjycb.20260929.004