Science map of academic misconduct

Main text

After a whistleblower exposed a large-scale fraud, there has been a surge in retractions, affecting scientific integrity and attracting extensive attention from various stakeholders.^1,2 Countries, institutes, and journals are among the entities receiving special attention.^3,4 However, our understanding of the actual prevalence of misconduct across disciplines and topics remains limited. Previous studies of academic misconduct have primarily focused on biomedical and life sciences,³ but have paid little attention to the broader scientific fields.⁴ In this research, a global analysis is conducted using data from 25,710 cases of academic misconduct retraction (AMR) out of 31,003 retracted articles between 2000 and 2023. The retracted articles and retraction notices were collected from journals indexed in the SCI, SSCI, and ESCI databases. They were then categorized into 10 macro-topics and 326 meso-topics using Citation Topics, which is a three-level hierarchical paper-level classification system available in the InCites database. Our findings suggest that the magnitude of academic misconduct varied widely across disciplines or topics, and the patterns of reasons for academic misconduct differed distinctly for each topic.

Misconduct occurs across all disciplines, with several topics being more severe

In the entire dataset, the AMR rate is relatively small, at 6.8 out of 10,000 articles. However, the AMR rate is heterogeneously distributed across disciplines, and it increases from 1.7 to 17.4 (Figure 1A). This indicates that academic misconduct is not equally prevalent across all disciplines. For example, in the discipline of Electrical Engineering, Electronics, and Computer Science (EE & Comp Sci), a total of 4,673 papers have been retracted due to academic misconduct, resulting in an AMR rate of up to 17.4, which is 10 times more than that of Physics. Additionally, Clinical and Life Sciences (Clin & Life Sci), which is known to be a reputed hotspot for academic misconduct, has the highest number of AMRs at 12,565, with an AMR rate of 8.9.

Figure 1.

Characteristics of academic misconduct retraction (AMR) across various disciplines or topics

(A) AMR rate per 10,000 papers for each discipline, indicating the total number of AMRs and the corresponding AMR rate.

(B) Science map globally depicting academic misconduct, with the top 10 topics labeled based on the number of retracted papers.

(C) AMR index for different meso-topics within various disciplines, with hollow circles representing the top 10 topics.

(D) Breakdown of the reasons for misconduct in the top 10 topics. The total number of papers considered is determined from journal papers (articles and reviews) indexed in the SCI, SSCI, and ESCI databases between 2000 and 2023.

At the meso-topic level, it has been determined that at least one instance of academic misconduct exists in 324 out of a total of 326 meso-topics. By mapping the retracted papers due to academic misconduct onto the science map, the layout of which is generated by VOSviewer to represent the citation network among meso-topics,⁵ a global representation can effectively highlight the meso-topics impacted by AMR (Figure 1B). In this map, each circle represents a meso-topic, and the size of the circle is directly proportional to the number of AMRs it contains. The range of the number of AMRs in the meso-topics varies significantly, from as low as 1 to as high as 2,105. The distance between circles reflects the similarity of topics measured by mutual citations. The labeled topics in the map indicate the top 10 topics with the highest number of AMRs. For instance, certain topics have a significantly large number of AMRs in the disciplines of Clin & Life Sci and EE & Comp Sci, such as Micro and Long Noncoding RNA (mlncRNA) with 2,105, Immunology with 447, Molecular and Cell Biology - Cancer, Autophagy, and Apoptosis (Mol & Cell Bio) with 678, Telecommunications with 940, Security Systems with 427, Computer Vision and Graphics (Comp Vis & Graph) with 633, Artificial Intelligence and Machine Learning (AI & ML) with 434, and so on. Interestingly, some topics exhibit relatively close citation distances, e.g., mlncRNA, Immunology, and Mol & Cell Bio.

The severity of misconduct varies greatly among different topics

Further analysis reveals a disproportionately high occurrence of academic misconduct in certain topics compared to their publication output. For instance, in the topic mlncRNA (Telecommunications), it contributed to 8.2% (3.7%) of the AMRs, despite accounting for only 0.4% (0.5%) of the published papers. Conversely, in the topics of Mathematics, Physics, and Chemistry, the proportion of retractions due to academic misconduct is significantly low compared to its publication output.

Hence, an AMR index has been proposed to assess the severity of academic misconduct in specific entities, such as topics, journals, institutions, and so on. It is defined as the ratio of the share of AMR to the share of published articles for a particular entity, regardless of the scale of retractions and articles. The AMR index of all meso-topics provides a macroscopic view of the severity of academic misconduct across disciplines or topics, as shown in Figure 1C.

It has been observed that out of a total of 326 meso-topics, 110 of them have an AMR index greater than 1, indicating a higher severity of academic misconduct compared to the overall baseline. Among these, 59 topics belong to the discipline of Clin & Life Sci. Notably, mlncRNA has been involved in the most severe academic misconduct issue, with the highest AMR index reaching 20.8. In the discipline of EE & Comp Sci, 22 out of 26 topics have an AMR index greater than 1. Interestingly, some topics with a smaller number of AMRs have higher AMR index values. For example, in the discipline of EE & Comp Sci, the topic of Remote Research and Education has the highest AMR index of 11.7, with 65 AMRs. Similarly, in the discipline of Social Sciences, the topic of Risk Assessment has the highest AMR index of 8.8, with 101 AMRs. On the contrary, almost all topics in the disciplines of Physics and Mathematics have an AMR index of less than 1, indicating a lower severity of academic misconduct in these topics.

Reasons for misconduct exhibit distinctive patterns among various topics

Consistent with earlier studies,³ 82.6% of articles were retracted due to various forms of academic misconduct. In recent years, however, the forms of academic misconduct have changed significantly.^1,2 Nearly a decade ago, academic misconduct was primarily attributed to occasional individual actions such as fabrication, falsification, plagiarism, duplication, and so on,² called traditional reasons in this study. However, in recent years, new forms of academic misconduct have become more prevalent, including Fake Peer-review, Paper Mill, and the emergence of artificial intelligence-generated content (AIGC).^1,2

Hence, 12,250 instances of academic misconduct were found to involve Fake Peer-review, accounting for 45.6% of all reasons for misconduct. Additionally, a significant number of retractions are also related to Paper Mill (8.5% or 2,275) or AIGC (4.4% or 1,170). On the contrary, traditional reasons account for 41.6% of all reasons. It should be noted that multiple reasons may be given for a retracted paper.

The reasons for misconduct in meso-topics exhibit more complex characteristics (Figure 1D). In the case of mlncRNA, the primary reason is Paper Mill, accounting for 48.7%, while traditional reasons also play a significant role, accounting for 38.9%. However, for the topics of Telecommunications, Security Systems, Comp Vis & Graph, AI & ML, and Edu & Educ Res, Fake Peer-review is the most common reason, accompanied by additional reasons such as AIGC and Paper Mill. In these topics, traditional reasons only contribute a small portion to the overall misconduct. In contrast, traditional reasons remain the main cause of misconduct in the topics of Phytochemicals, Anesthesiology, Immunology, and Mol & Cell Bio.

Conclusion and discussion

Although retraction is a scientific self-correction mechanism, the recent surge in the number of retractions is a concerning trend. This study reveals the widespread occurrence of academic misconduct across various topics, but the severity of misconduct varies significantly among them. Through the AMR index, it was discovered that certain topics face particularly severe issues of academic misconduct, emphasizing the urgent need for increased attention and efforts to address these problems. The emergence of large-scale fraud has further complicated and obscured the issue of academic misconduct. Our results suggest that interventions for addressing academic misconduct should be discussed on a discipline- and topic-specific basis while considering their characteristics. We hope that this study can provide valuable insights and inspiration for effectively managing and addressing the issue of academic misconduct.

Published Online: February 19, 2024