The retraction gender gap: Are mixed teams more vulnerable? Open Access

https://www.webofscience.com/api/gateway/wos/peer-review/10.1162/qss_a_00353

Gender diversity is increasingly acknowledged as an important factor influencing not only the inclusiveness and fairness of research, but also the innovation and creativity that drive scientific progress (Krammer & Dahlin, 2024; Maddi & Gingras, 2021; Nielsen, Bloch, & Schiebinger, 2018). This issue of gender diversity arises at several interrelated levels: at the level of scientific careers—although women make up a significant proportion of the scientific community, they remain significantly underrepresented in higher academic ranks and leadership positions; at the level of scientific disciplines and epistemic cultures: mathematics, physics, engineering sciences are characterized by a marked underrepresentation of women; and, of course, at the level of scientific collaboration and authorship: as Sugimoto and Larivière (2023, p. 48) pointed out, “overall, women’s authorships are more likely to be in collaboration than men’s and (…) women are less like to be sole author in domains where single authorship is particularly distinctive (…)”.

The issue of gender diversity in scientific research is a global challenge and despite the major efforts made over many years to understand the mechanisms associated with the production of these differences (see for example Zuckerman, Cole, & Bruer, 1991), considerable work remains to achieve true gender equity in scientific leadership and output. Addressing these situations of inequity requires not only identifying their multiple origins but also implementing policies that promote equitable collaborations and transparency in academic labor, encouraging scientific societies and institutions to regularly assess and improve the diversity of their membership and leadership.

While the benefits of gender diversity for innovation and research productivity are well documented, its potential impact on the reliability of scientific publications is less explored. This highlights an important yet insufficiently investigated question: Does gender diversity help mitigate the potential problems in scientific publications, ranging from simple errors to serious misconduct, which often lead to retractions? Retraction, a key indicator of how well science is (not) doing, occurs when it is shown that a publication contains serious errors or that its author has committed scientific misconduct or ethical violations. It is intended to correct the scientific record (Wray & Andersen, 2018). Understanding whether gender-diverse research teams are less prone to these problems could provide valuable insights into how to promote scientific reliability.

In this article, the general issue of the relationship between gender diversity and the reliability of science is broken down into two research questions: (a) Does gender diversity in research teams have an impact on article retraction? (b) Are there discernible disparities in the reasons underlying the retraction, depending on the gender composition of scientific publications? For each of these questions, this article provides answers based on the processing and analysis of a random sample of one million publications from OpenAlex, covering 2,645,304 authors.

Throughout the 20th century, the presence of women in science grew steadily. Although the number of women has increased in all scientific disciplines (Rossiter, 1997, 2012), numerous studies in the sociology of science have highlighted the persistence of gender disparities across many dimensions of the academic career (salaries, career achievement, prestigious status or awards, etc.) (Larivière, Ni et al., 2013; Lincoln, Pincus et al., 2012; Rossiter, 1993). Moreover, these disparities between men and women age peers widen between the beginning and end of their careers (Sugimoto & Larivière, 2023; Zuckerman et al., 1991).

Since the 1980s, quantitative analyses have repeatedly shown that women’s scientific productivity is generally lower than that of men (Cole & Zuckerman, 2024; Reskin, 1978; Tower, Plummer, & Ridgewell, 2007). To quote some recent results, except in a few highly feminized disciplines (arts and health) where the gap between men’s and women’s publications is small, men publish 20% more than women across all disciplines, with the gap reaching a third (mathematics, physics, social sciences) and more (50% in psychology) (Sugimoto & Larivière, 2023). This trend of studies has especially emphasis the link between gender gap in access to a range of gains in academia and gender disparities in scientific production and citation insofar as authorship plays a key role in access to scientific advantages (position, status, funding, prizes, etc.).

Sugimoto and Larivière (2023), relying on scientometric analyses of millions of publications spanning all disciplines, have made a significant contribution to our understanding of the mechanisms that hinder the development of women’s careers, and that are largely rooted in the functioning of authorship attribution. Paying attention to the distribution of authorship in publications, they highlight a systematic disadvantage of women, who are less likely to occupy positions of value in bylines. They show that women are less likely than men to write as a sole author, whereas signing alone is particularly rewarding. If single-authored articles have largely decreased with the rise of collective authorship in all disciplines since the Second World War, single authorship reflects a type of intellectual labor considered of high scientific value, not to mention the high citation rate that sole-authored articles usually capture.

If they publish less as single authors, they are more likely than men to write in collaboration. But in all fields, it is men who occupy the dominant and therefore rewarding positions in the byline (i.e., first and last author). Women more often tend to occupy positions in the byline that mark a junior status or a low position in their career, or that indicate the performance of technical tasks of relatively low prestige (i.e., early and midpositions). However, they are more likely to occupy a leadership role by signing as first author if the size of the team increases. However, above a certain team size, they tend to lose this dominant position, a trend reflected in the analyses based on our data.

Numerous studies have delved into the realm of retracted scientific articles, exploring the multifaceted aspects surrounding them (Fang & Casadevall, 2011; Wray & Andersen, 2018). These investigations have examined diverse angles, encompassing the reasons for retractions, ethical considerations, methodological flaws, and the potential consequences for scientific trust and credibility (Cokol, Ozbay, & Rodriguez-Esteban, 2008; Pfeifer & Snodgrass, 1990; Steen, 2011b; Steen, Casadevall, & Fang, 2016).

Gender disparities in science extend beyond authorship and career advancement; they also manifest themselves in areas such as the correction and retraction of scientific work. The existing literature has shown that there are notable gender differences in acknowledging errors, recanting, and correcting scientific findings. Studies have shown that female scientists are more likely to backtrack on their research or admit mistakes than their male counterparts (Wadman, 2005). This gender disparity in correctional practices can be attributed to a variety of factors including societal expectations, biases, and cultural norms that put additional pressure on women to uphold high standards of accuracy and integrity (Kaatz, Vogelman, & Carnes, 2013). Additionally, systemic biases and gender inequalities within the scientific community can influence the reception and response to errors, leading to differential treatment based on gender.

Gender differences in scientific misconduct is a topic that has gained increasing attention in the recent scientific literature, with several studies shedding light on various aspects of this issue. Fang, Bennett, and Casadevall (2013) conducted a comprehensive analysis of the Office of Research Integrity (ORI) database, examining 228 cases of scientific misconduct. Their findings revealed that a significant majority of these cases involved male researchers, particularly professors. However, it is important to point out that these findings do not imply that male researchers are inherently less ethical or have less integrity than their female counterparts. Therefore, one needs to investigate the historical and sociological factors contributing to these disparities, as underscored by Kaatz et al. (2013).

Examining the presence of women in retracted publications, Decullier and Maisonneuve (2023) conducted one of the first studies on this topic. Their analysis of 113 retracted publications revealed that 37.2% of them were authored by women, either as the first author or as sole authors. Notably, fraud and plagiarism accounted for 28.6% and 59.2% of retractions, respectively, for both men and women. In the field of biomedical sciences, Pinho-Gomes, Hockham, and Woodward (2023) examined gender differences in the authorship of retracted articles, utilizing data from Retraction Watch. Among 35,635 retracted biomedical articles spanning from 1970 to 2022, women accounted for 27.4% of first authors and 23.5% of last authors. The underrepresentation of women was particularly pronounced in cases of fraud (approximately 19%) (see also Sebo, Schwarz et al., 2023). Notably, the highest representation of women as first authors was observed in cases related to editorial issues (35%) and errors (29.5%). Overall, most retractions (60.9%) involved male researchers as both first and last authors. The authors suggested that achieving gender equality could contribute to enhancing research integrity in the field of biomedical sciences.

In a study by Ribeiro, Mena-Chalco et al. (2023), the authors investigated the behavior of self-retraction due to honest errors, considering factors such as country, research domain, and gender. Analyzing 2,906 retracted publications categorized under “errors” between 2010 and 2021, the study revealed that women scientists accounted for 25% of self-retractions due to honest errors. However, further research is warranted to develop standardized indicators that measure the proportional representation of women in science, specifically in the context of retractions. Furthermore, Satalkar and Shaw (2019) conducted interviews with 33 Swiss researchers to explore the early influences on concepts of honesty, integrity, and fairness in research. The findings highlighted the significance of early education, moral values instilled by families, and participation in team sports as the primary influencers. Notably, two-thirds of the participants had not received any formal training on research integrity, which may increase the likelihood of honest errors due to a lack of knowledge.

While gender disparities in retractions are evident in the literature, it is important to interpret these findings in light of broader structural and historical factors. The overrepresentation of men in retractions aligns with their historically higher representation in academia and greater overall scientific productivity. However, as highlighted by multiple studies, this does not imply inherent differences in ethical behavior or research integrity between genders. Instead, it underscores the complex interplay of systemic factors, social expectations, and disciplinary norms.

Previous research has often focused on misconduct, ethical violations, or errors leading to retractions, but less attention has been given to how the composition of research teams might influence these outcomes. This study seeks to bridge this gap by examining the role of gender diversity within research teams and exploring whether different gender balance led to distinct patterns in retraction reasons. To achieve these objectives, we articulate two pivotal research questions guiding our exploration:

• RQ 1: Does gender diversity in research teams have an impact on article retraction?

• RQ 2: Are there discernible disparities in the reasons underlying the retraction, depending on the gender composition of authorship in scientific publications?

Figure 1 provides an overall summary of the methodology used in this study, linking the different data sets and analyses conducted. The diagram illustrates two main phases of the analysis. In the first phase, we explored the gender composition of publications from the Retraction Watch database and compared it to a random sample drawn from OpenAlex. This phase also included an investigation of the reasons for retractions, with a particular focus on male–female collaborations. In the second phase, we delved deeper into the reasons for retractions using a logistic regression analysis to identify the main determinants, incorporating gender and other collaborative factors.

The first phase began with the validation and processing of data from OpenAlex (693,621 publications with identifiable authors) and Retraction Watch (∼40,000 retracted publications). A random sample of one million publications was drawn from OpenAlex, followed by a gender identification process for the authors. The retracted publications from Retraction Watch were then matched with their corresponding metadata in OpenAlex via DOIs, resulting in a refined subset of 15,869 retracted publications. This subset was balanced with a random sample of 15,869 nonretracted publications to ensure robust analysis and avoid biases stemming from class imbalances.

The second phase involved a multivariate analysis conducted on this balanced data set of 31,738 publications (retracted and nonretracted). This analysis employed logistic regression to evaluate the relationships between retractions and various factors, including the gender composition of collaborations. Robustness checks were performed to validate the results and ensure their reliability across different model specifications (see the Supplementary material).

The data for scientific publications were extracted from the OpenAlex database, a rich and inclusive source of academic literature (Priem, Piwowar, & Orr, 2022). We chose to work with a random sample of one million publications to ensure sufficient diversity and representativeness. After an initial selection, we retained 990,100 publications with metadata on the authors, encompassing a total of 2,645,304 authors.

For our analysis, we focused on publications where we were able to identify the gender of at least some of the authors. This approach aligns with the methodology used in Larivière et al. (2013), where publications were included even if the gender of all authors could not be identified. This allowed us to reduce our sample to 693,621 publications, authored by 1,758,197 identified authors, representing 70% of the initially selected publications (990,100).

OpenAlex topics are based on an advanced method of classifying research works according to their content. The topics are structured in a four-level hierarchy and cover approximately 4,500 distinct subjects. For our analyses, we used the intermediate level, comprising 26 disciplines, to determine the average proportion of publications by discipline. Furthermore, we employed an aggregated level of five major categories as a control variable in the regression. For more details on the topics, refer to the following link: https://docs.openalex.org/api-entities/topics/get-lists-of-topics.

The data on retracted publications were kindly provided by the Center for Scientific Integrity in April 2023 (https://retractionwatch.com/). The database contains nearly 40,000 retracted publications. In addition to information regarding the retraction status of publications, we also utilized the reasons for retraction to analyze gender differences. To accomplish this, we calculated a double ratio as follows.

The double ratio for retraction reasons is calculated by comparing the proportion of retractions attributed to a specific reason in publications led by men (including those with a sole male as author, multiple male authors, and mixed-gender publications where the first author is male) to the proportion of that reason across all publications in the Retraction Watch database. Similarly, we applied this calculation to publications led by women (including those with a sole female as author, multiple female authors, and mixed-gender publications where the first author is female). This ratio helps determine if a retraction reason is more frequently associated with publications led by men or women compared to all publications. A double ratio greater than 1 indicates a disproportionate prevalence of that retraction reason among publications led by men or women, respectively.

To avoid arbitrary choices of retraction reasons, we employed a fractional counting approach, considering all reasons. For instance, if a publication was retracted for two reasons, A and B, we assigned a count of 0.5 to each reason. This approach has two advantages: First, it allows us to measure the relative intensity of each retraction reason by type of collaboration; second, it enables us to sum the counts so that the total count for all reasons and collaboration types equals the total number of retractions (i.e., 36,297).

To measure the relative weight of different reasons according to the type of men–women collaboration, we applied the method of calculating the activity index traditionally used in bibliometrics to measure disciplinary specialization. In our case, this indicator is calculated as follows. For a given retraction reason, such as “Misconduct by Author,” the index is calculated by comparing the proportion of “Misconduct by Author” within publications “led by men” (or by women) to the proportion of the same reason in the entire Retraction Watch database. The indicator varies around 1. Thus, an index of 1.16 means that “Misconduct by Author” is on average 16% more prevalent in publications led by men compared to the overall retractions. Next, we subtracted the index for the group of publications led by women from the index for the group of publications led by men (and vice versa). This allows us to obtain the percentage difference between the two for each reason.

The purpose of this section is to provide details on the disciplinary composition of the OpenAlex database, which is used as a benchmark for Retraction Watch. While the primary focus is on gender identification, we also present the distribution of publications across different disciplines (topics) in OpenAlex. This serves to validate and functionally test known trends in the existing literature, such as the underrepresentation of women in mathematics and the higher representation of women in health professions.

To identify the gender of authors, several methods exist (Larivière et al., 2013; Wais, 2016; West, Jacquet et al., 2013). In this paper, we opted to utilize Wais’s (2016) method and the R package “genderize.io.”

Table 1 illustrates the distribution of author genders within the OpenAlex data set. Of the total 2,645,304 authors analyzed, 46% were identified as male and 21% as female. A significant portion, 17%, remained undefined, while 15% were represented by initials. An additional 2% were classified as unisex. This breakdown underscores the consistency of findings with prior research on gender representation in scientific literature (see for example Larivière et al., 2013).

Figure 2 shows the average proportion of women per article in different disciplines. It reveals that the overall average across disciplines is 31%, but also highlights significant disparities in women’s participation in scientific research. This uneven distribution raises questions about the multiple factors that influence the representation of women in different disciplines.

Health professions and social sciences stand out for their high female participation. Leading the way, health professions have a proportion of 45.3% of women per article, closely followed by psychology at 44.9% and nursing at 40.6%. Additionally, the social sciences, with an average proportion of 40.2%, reflect women’s presence in disciplines focusing on human behavior and social dynamics.

The arts and humanities (37.7%), as well as veterinary sciences (36.4%), also show significant female participation, higher than the overall average across disciplines (31%). Fields such as dentistry (35.2%) and immunology and microbiology (34.9%) indicate that women are beginning to carve out a place in biomedical sciences.

In contrast, physical, material, and technical sciences show much lower female participation: Earth and planetary sciences (23.7%), chemistry (23.7%), and materials science (21.7%). These sectors have traditionally been perceived as male domains, which continues to affect female participation despite efforts to promote gender equality. Chemical engineering (21.5%), general engineering (19.6%), and mathematics (19.6%) follow this trend, highlighting ongoing challenges in attracting and retaining women. Finally, physics and astronomy display the lowest proportion of women per article, with only 16.2%.

Table 2 provides an overview of the variables used in the logistic regression analysis on the retractions of scientific publications. This analysis aims to identify the factors associated with the likelihood of a publication being retracted. The explanatory variables include classifications based on the gender of the leading authors, the open access status of publications, the number of citations received, research funding through grants, and the scientific disciplines involved. The dependent variable is binary and indicates whether a publication has been retracted or not. This table offers a detailed view of the categories and levels of each variable, thus enhancing the understanding of the factors considered in our analysis model.

We used a logistic regression model to analyze the probability of a publication being retracted (dependent variable: retracted or not). All explanatory variables, as listed in Table 2, were entered simultaneously into the model without using any stepwise selection procedure.

The results of the multivariate logistic regression analysis are presented in Figure 3, where the effect of each variable was adjusted for all other parameters included in the model.

In this section, we present the main results of our analysis. First, we examine the distribution of publications based on the type of men-women composition in the overall Retraction Watch database and within the OpenAlex random sample. This comparison reveals significant differences in gender representation and collaboration dynamics between retracted articles and those from the general literature. Second, we present the distribution based on the retraction reasons, focused on the most reasons for both publications where men are leader, and for that where women are leader, respectively. Finally, we discuss the results of the logistic regression.

To analyze the gender composition of retracted publications, we inferred the gender of authors separately for both the Retraction Watch database and a random sample extracted from OpenAlex. Since extracting the entirety of the OpenAlex database was not technically feasible due to capacity limitations, we opted for a large random sample comprising 693,621 scientific publications with identified authors. This sample size was chosen based on the law of large numbers, under the assumption that the gender distribution within this sample would converge to the overall gender distribution in OpenAlex. This large-scale benchmark serves as a baseline for comparison, enabling us to evaluate whether the gender composition of retracted publications in the Retraction Watch database deviates from the expected distribution observed in the broader scientific publication landscape.

Table 3 compares the distribution of genders and types of collaboration in the OpenAlex database, which contains a random sample of scientific publications, and Retraction Watch, which specifically focuses on retracted articles. The table shows that collaborations composed exclusively of men are slightly more prevalent in Retraction Watch (31%) compared to OpenAlex (28%). This difference may suggest that retracted articles are slightly more likely to come from collaborations led by men.

The proportion of articles written by men alone is significantly lower in Retraction Watch (11%) compared to OpenAlex (24%). This disparity could indicate that articles authored solely by men are less frequently retracted. Research conducted individually may be less susceptible to complex collaborations, thereby reducing the risk of errors or fraud that could lead to retractions. This trend could also reflect a tendency for both men and women working alone to be more cautious or autonomous in their research, thereby minimizing the risks of retraction. Similar to men, articles authored solely by women represent a much smaller proportion in Retraction Watch (5.3%) compared to OpenAlex (12%).

Collaborations between men and women where a man is the first author are significantly more frequent in Retraction Watch (25%) compared to OpenAlex (13%). This overrepresentation suggests that articles where a man is the first author in mixed-gender collaborations are more likely to be retracted. Collaborations between men and women where a woman is the first author are also more frequent in Retraction Watch (17%) compared to OpenAlex (12%). While this difference is less pronounced than for male first authors, it suggests that articles where women are first authors in mixed-gender collaborations are also more likely to be retracted. The proportion of exclusively female collaborations is similar between the two databases, with 6.6% in OpenAlex and 6.2% in Retraction Watch.

The analysis of gender distribution and collaboration types between retracted articles and those from the general database reveals significant differences. Collaborations led by men, particularly those where men are first authors, are more likely to appear in Retraction Watch. In contrast, women working alone or in exclusively female groups are less represented in retracted articles, suggesting potential differences in terms of methods or approaches.

However, it is crucial to note that controlling for factors that may affect these distributions, such as discipline, publication openness status, funding, or academic impact, is necessary. This will be addressed further in the logistic regression analysis later in this article.

The double ratios provided in Figure 3 indicate significant variations in the frequency of retractions based on gender and collaboration configurations¹. Collaborations where a man is the first author in mixed-gender teams (Men-Women ∣ M first) have a double ratio of 1.92, meaning they are 92% more frequent among retracted articles compared to the general data set. Similarly, collaborations where a woman is the first author in mixed-gender teams (Men-Women ∣ W first) show a double ratio of 1.42, indicating a 42% increase in the frequency of retractions for these articles.

Conversely, articles authored solely by men (Men only) have a double ratio of 1.11, suggesting a slight overrepresentation among retractions compared to their proportion in the general data set. In contrast, articles authored solely by women (Women only) exhibit a double ratio of 0.94, indicating a slight underrepresentation among retractions, suggesting a slightly lower risk of retraction for these articles. Research conducted independently by men alone (Man alone) shows a double ratio of 0.46, while research conducted by women alone (Woman alone) has a double ratio of 0.44. These results indicate that individually authored articles, whether by men or women, present a significantly reduced risk of retraction.

To draw robust conclusions about the gender impact on retractions, it is crucial to consider disciplinary specifics, research practices, and other relevant contextual variables. Double ratios provide a precise comparative view of trends, but thorough analysis is necessary to interpret these results and avoid premature generalizations.

In Tables 4 and 5, we display only the top reasons, in Retraction Watch database, for publications led by men and for publications led by women, respectively. In other words, Table 4 shows the main reasons for retraction for publications led by men, while Table 5 shows the main reasons for retraction for publications led by women. For example, notably, the reason "Misconduct by Author" is 54% more prevalent in publications led by men than in publications led by women. The last column in the two tables indicates the corresponding number (fractional count: see Section 4.3) of concerned publications for a given reason.

Table 4 compares the most common retraction reasons between publications led by men (Men-Women ∣ M first, Men only, Man alone) and those led by women (Men-Women ∣ W first, Women only, Woman alone). It shows significant differences in retraction motifs, expressed through ratios. These ratios indicate the relative frequency of each retraction reason among men compared to women, providing insights into the dynamics of retraction based on the gender of the lead author. The colors represent those of a heatmap: The bluer it is, the lower the double ratio; conversely, the redder it is, the higher the double ratio.

The findings show that several retraction reasons are more prevalent in publications led by men. For instance, lack of approval from the company or institution is 76% more frequent in publications led by men compared with those led by women. This suggests that compliance with ethical regulation such as IRB may have a gender dimension. Additionally, plagiarism of images is 18% more common in publications led by men. This reason indicates improper use of images from other sources without proper citation, raising questions about diligence in verification and the credibility of visual data included in publications led by men.

False authorship, where authors’ names are fraudulently used during manuscript submission, is also more frequent in publications led by men, with a ratio of 1.17 compared to 0.61 in publications led by women. This practice may be associated with increased pressure to enhance the visibility or impact of publications, often at the expense of academic and ethical integrity. Furthermore, the results show that publications led by men are more frequently retracted due to misconduct by authors, such as data falsification or manipulation of results. These reasons are respectively 16% and 15% more common among men compared to women, highlighting potential differences in perceived ethical standards or research management practices.

The analysis of the retraction double ratios, as presented in Table 5, reveals significant differences between the reasons for retraction for publications led by women compared to those led by men. The reasons for which women are more frequently affected indicate a tendency where female-led publications are retracted primarily for errors or procedural violations rather than for serious offenses like plagiarism or fraud. As in Table 4, the colors represent those of a heatmap: The bluer it is, the lower the double ratio; conversely, the redder it is, the higher the double ratio.

Women are more often retracted for reasons that include textual errors, updating issues, and administrative complications. For instance, the reason “Withdrawn (out of date)” has a double ratio of 1.50 for women, which is 72% higher than for men, who have a ratio of 0.78. This suggests that articles led by women are more likely to be retracted for reasons related to the updating of guidelines or professional reviews. Upon closer examination of publications retracted for this reason (“Withdrawn (out of date)”), all of them (361 publications) are related to the medical field, with a majority involving Cochrane, an internationally recognized organization for systematic reviews, mainly in the health domain.

Another common reason for retraction for women is related to the reuse of content from dissertations or theses, with a ratio of 1.45 for women compared to 0.81 for men. This shows that women are more often retracted for having included elements from their own previous academic work, possibly without adhering to citation requirements or content reuse regulations.

Women are also more likely to be retracted for “Error in text” (1.39 compared to 0.83 for men) and for “Retract and replace” processes (1.35 compared to 0.85). These results indicate that publications led by women more frequently encounter issues of precision in the textual content and the need for substantial corrections that lead to retraction and republication. Reasons related to textual plagiarism and copyright violations are also more common among publications led by women. Textual plagiarism shows a ratio of 1.30 for women, in contrast to 0.87 for men. This indicates that female-led publications are more often retracted for using text sections without appropriate citations. Additionally, retractions for “Euphemisms for plagiarism” and “Copyright claims” are also more frequent for publications led by women, with respective ratios of 1.25 and 1.22.

Globally, as shown in Tables 4 and 5, retraction reasons can differ significantly between publications led by men and those led by women. Men’s publications are more frequently retracted for serious ethical violations, such as data falsification, manipulation of results, and plagiarism, including image plagiarism. These infractions often point to issues related to research integrity and misconduct. Men are also more likely to face retractions due to institutional or regulatory noncompliance, such as lack of approval from a company or ethics board, highlighting potential gaps in adherence to research governance.

In contrast, women’s publications tend to be retracted for reasons related to procedural errors, such as textual mistakes, the reuse of content from previous academic work, and issues with updating guidelines in rapidly evolving fields like medicine. These retractions are generally less about intentional misconduct and more about technical or methodological oversights. This suggests that female-led publications may be scrutinized more for precision and adherence to formal processes, while male-led publications may face more serious retractions due to ethical breaches or misconduct. Overall, this points to a gendered dimension in how retractions are triggered and managed across academic publications.

Our analysis so far has described significant disparities between women and men, but how can we explain them? What factors are at work? This section presents the results of the logistic regression analysis on the key factors influencing the likelihood of scientific publication retractions.

Figure 4 illustrates the odds ratios (OR) for various explanatory variables, including gender composition of the author team, open-access status, journal impact, research funding, and scientific discipline. Each variable’s impact on the probability of retraction is depicted, with odds ratios greater than one indicating an increased likelihood of retraction and values less than one indicating a reduced likelihood. Confidence intervals are shown to provide insights into the statistical significance and reliability of the estimates. This figure helps to highlight the key determinants and the magnitude of their effects on publication retraction probabilities, offering a clearer understanding of the dynamics at play.

The results from the regression analysis provide an interesting insight into the factors associated with the risk of retraction of scientific publications. These findings highlight the complex influences of gender, research team size, funding, discipline, open access, and journal impact on the likelihood of a publication being retracted. These findings may appear counterintuitive. The interpretative paths developed in this section must be approached with caution and should be tested on the basis of dedicated empirical investigations.

The results show significant differences based on gender and team composition. Publications by mixed-gender teams, consisting of both men and women, are more likely to be retracted than those of teams composed only of men or women. All-female teams have a slightly lower risk of retraction compared to mixed teams, while individual authors, regardless of gender, are significantly less likely to have their publications retracted.

This dynamic can be partially explained by the management of roles within mixed teams. Studies on scientific collaboration indicate that gender diversity can lead to increased creativity and innovation but can also result in tensions or coordination challenges, potentially related to differing expectations or social norms. These differences could lead to errors or misunderstandings that heighten the risk of retraction. Unisexual teams, on the other hand, might be more homogeneous in their communication and working styles, facilitating coordination and decision-making, thereby reducing retraction risks.

The fact that individual authors are less likely to be retracted may seem counterintuitive, given that individual work may be subject to less internal oversight. One explanation for this result may lie in the fact that papers written by a single author are more likely to be review papers or theoretical papers (Sugimoto & Larivière, 2023). They are therefore different in nature from those written by teams collaborating on more experimental research programs. These papers are also often considered to be more distinctive, which undoubtedly implies greater vigilance in view of the value attributed to them.

The results also indicate that the size of research teams affects the risk of retraction. Generally, a larger number of authors is associated with a lower risk of retraction. This may seem counterintuitive at first, as one might assume that having more authors increases the risk of coordination issues or errors. However, larger teams often have more formalized structures, with clear division of labor and distribution of responsibilities. Tasks are often specialized, allowing for a more even distribution of workloads, verification of contributions, and improved quality control prior to publication.

However, when introducing a specific variable for medium-sized teams (between three and 10 authors), it is observed that these teams have a higher risk of retraction. This could reflect a less formal structure or a lack of clarity in the distribution of roles within these teams. Unlike larger teams, where coordination and supervision are often organized by subgroups, medium-sized teams may suffer from ambiguity in responsibilities and a lack of hierarchy. This lack of structure can contribute to undetected errors or shortcomings in reviewing and validating results, thereby increasing the risk of retraction.

Funding appears to play a major protective role against retractions. Publications resulting from funded research are 93% less likely to be retracted compared to unfunded publications. This finding aligns with the notion that funded studies benefit from more significant resources, enabling stricter quality control at every stage of the research. Funded projects often have reporting obligations and increased transparency requirements toward funding agencies, which may encourage researchers to adhere to rigorous scientific practices.

Funded studies also have access to more sophisticated methodological tools and technical support, which can enhance the quality of research. Furthermore, funding agencies often impose higher standards in terms of ethics and transparency, which can also contribute to reducing the risk of retraction. This result is particularly relevant in a context where research funding is becoming increasingly competitive, and the quality of publications may serve as a key indicator for obtaining new funding.

Differences in retraction rates by scientific discipline are also notable. Compared to life sciences, which serve as the reference category in the analysis, social sciences and physical sciences exhibit significantly lower retraction rates. This may be attributed to distinct research practices and disciplinary norms.

Physical sciences impose strict standards for verifiability and reproducibility of experiments, with a strong propensity for publishing preprints. This may contribute to lower retraction rates, as errors are more likely to be detected before publication. Like the physical sciences, the social sciences have a low retraction rate, but probably for different reasons: The singularity of the data produced by the nonexperimental sciences, which make up a large part of the social sciences, makes it difficult to exercise the verifiability essential for retraction. Added to this is the limited use of postevaluation practices.

The results show that open access publications have a 4.31 times higher likelihood of being retracted than nonopen access publications. This finding can largely be explained by the increased visibility of openly accessible publications. Being more readily accessible, these works are subject to more rigorous scrutiny by the scientific community, increasing the chances of detecting errors or fraud postpublication.

This may also reflect differences in peer review processes between open access journals and those following traditional models. Some open access journals, particularly those with low publication fees, have been criticized for having less rigorous peer review processes. However, it is essential to note that open access itself should not be viewed as a direct cause of retraction but rather as a model that, due to its transparency and increased visibility, is subject to a higher level of scrutiny. Therefore, to send a positive signal of rigor and transparency to the scientific community, open access journals may implement high-quality control measures and engage in more retractions.

Publications in high-impact journals are also more likely to be retracted. This finding aligns with existing literature, which shows that prestigious and highly visible journals are scrutinized more closely by the scientific community. Results published in these journals often have significant implications for their respective research fields, attracting a larger number of critical readers. Consequently, errors or fraud in these publications are more likely to be discovered and reported.

Several studies have shown that journal retraction rates are correlated with the impact factor (Cokol, Iossifov et al., 2007; Fang & Casadevall, 2011; Steen, 2011a), and even more so with the h-index (Nagella & Madhugiri, 2020), and that, consequently, retraction rates are higher for highly cited journals.

To sum up, these results illuminate various factors influencing the risk of retraction of scientific publications. Whether concerning gender, team size, funding, discipline, open access, or journal visibility, these factors interact in complex ways to determine the likelihood of a publication being retracted. These findings provide a valuable foundation for understanding the mechanisms of quality control in research and identifying avenues for improvement, particularly regarding the transparency of editorial processes and the management of research teams.

This study aims to analyze the impact of gender diversity in research teams on the likelihood of retraction. The findings highlight the complex dynamics that contribute to the retraction of scientific articles, providing a basis for understanding how multiauthor collaborations and individual contributions differ in terms of susceptibility to retraction.

Multiauthor publications, particularly those involving mixed-gender dynamics, show an increased likelihood of retraction. This elevated probability could be attributed to the complexities inherent in managing contributions from multiple authors. Collaborations involving diverse disciplinary backgrounds, varying levels of expertise, and different interpretations of ethical standards can lead to discrepancies that may result in retractions. These findings align with previous research suggesting that the risk of errors and ethical violations increases with the number of authors involved, due to the heightened complexity of ensuring data integrity and maintaining consistent standards within a collaborative team.

An interesting aspect is that publications with a male as the first author in a mixed-gender collaboration have a higher likelihood of retraction compared with those with a female as the first author. This might indicate that traditional roles and expectations associated with gender in scientific collaborations influence the dynamics of these projects. Men, often perceived as taking a leadership role in collaborative efforts, may be subject to more rigorous scrutiny, potentially contributing to a higher probability of retraction. However, this observation requires further investigation to comprehensively understand the underlying causes.

In contrast to the trends observed in multiauthor collaborations, the gender of the author in single-author publications does not have a significant impact on the likelihood of retraction. This suggests that responsibility and scrutiny in single-author publications are clearer, likely resulting in more rigorous attention to detail and better adherence to ethical standards, regardless of the author’s gender. The clarity of responsibility in single-author publications may lead to fewer opportunities for errors or misconduct to go unnoticed, thus reducing the overall probability of retraction.

Additionally, the practices and policies of open-access (OA) journals might contribute to the higher retraction rates observed. OA journals, which are often newer and more recent compared to traditional subscription-based journals, may place a stronger emphasis on maintaining high standards and protecting their reputation to establish credibility in the scientific community. As part of their quality assurance practices, these journals may employ more stringent peer-review processes and have robust mechanisms in place to detect ethical violations. This proactive approach to identifying and retracting problematic articles ensures that any issues are promptly addressed, contributing to the higher retraction rates observed in OA publications. Furthermore, the increased visibility and accessibility of OA articles make them more likely to be scrutinized by a wider and more diverse audience, further increasing the likelihood of detecting errors or misconduct. Thus, the positive impact of OA on retractions reflects both increased scrutiny and a commitment to maintaining the integrity of published research. Consequently, it is important to note that open access in itself does not necessarily imply lower quality; on the contrary, the higher retraction rate could reflect a greater opportunity for error detection by a wider and more diverse audience.

Furthermore, the positive association between the journal impact and the likelihood of retraction underscores the role of attention and scrutiny in maintaining scientific integrity. Highly cited articles, being more visible and influential, attract more critiques, increasing the chances that underlying issues are detected and addressed through retraction if necessary. This finding suggests that the scientific community is actively engaged in quality control of highly influential research, which is crucial for maintaining the credibility and reliability of scientific literature. Consequently, this link can be interpreted as a consequence of the fact that highly cited articles attract more attention and scrutiny, which increases the likelihood that issues will be detected (Nagella & Madhugiri, 2020), and also that high impact journals are more likely to trigger replication, which leads to more systematic detection of poorly reproducible or fraudulent results. Finally, the policies of high-impact journals are geared towards the publication of important results and are of great importance for the careers of scientists, which tends to encourage fraud. Indeed, Steen (2011a) notably showed that compared to authors who produce erroneous articles, authors who produce fraudulent articles specifically target journals with a high impact factor.

Moreover, our findings align with previous research, confirming that larger research groups generally have a lower probability of retraction, as demonstrated by Rathmann and Rauhut (2019), who suggested that team size may facilitate better social control and oversight. Similarly, Sharma (2021) highlighted that smaller teams are more susceptible to retractions, reinforcing the notion that team size plays an important role in research integrity. However, our analysis introduces a nuance regarding medium-sized teams (three to 10 authors), which are found to be at a higher risk of retraction compared to larger teams. This observation suggests that while collaboration can mitigate misconduct risks, medium-sized teams may struggle with ambiguous roles and responsibilities, potentially leading to oversight issues that larger, more structured teams can avoid. Furthermore, the findings of (Tang, Hu et al., 2020) reinforce this complexity, as they noted that collaboration does not inherently increase retraction likelihood, despite a higher prevalence of multiauthored retracted articles.

Finally, our results highlighted that publications led by men are more often retracted for severe reasons, like misconduct, while those led by women are more often retracted for reasons that include textual errors, updating issues, and administrative complications. For instance, the reason “Withdrawn (out of date)” is 72% higher than for men. It seems important here to focus on this point to highlight the importance of retracting outdated studies in the medical field and the contrast with other disciplines. Journals like Cochrane, which often include meta-analyses, aim to synthesize the available scientific evidence on the effectiveness of medical treatments or health interventions, whether involving drugs, therapies, or medical practices. Unlike original studies that focus on specific molecules or therapies, Cochrane compiles and analyzes the results of multiple published studies to produce a comprehensive summary on a given topic.

The retraction or withdrawal of outdated articles is generally more prevalent in the medical field, especially with systematic reviews and meta-analyses. Several reasons explain why this practice is more common in health research compared to other fields:

• Direct impact on healthcare: In the medical field, scientific publications directly influence clinical practice, public health policies, and patient treatment decisions. If a systematic review or meta-analysis contains errors or becomes outdated due to new evidence, it can have serious consequences. Clinicians often rely on the recommendations from these studies to prescribe treatments, so if this information is incorrect or outdated, it can endanger patients’ health (Brignardello-Petersen, Carrasco-Labra, & Guyatt, 2021; Grimshaw & Russell, 1993).

• Evolving nature of medical evidence: Medicine is one of the fields where knowledge progresses rapidly. New clinical studies can invalidate or refine the conclusions of earlier studies (Faber, Eriksen, & Hammer, 2023). Reviews that produce meta-analyses are designed to provide a synthesis of the best available evidence at a given time, but this evidence may change over time. When significant new data emerge, a review may be retracted or updated to reflect the new knowledge. In other fields, such as the humanities or social sciences, knowledge changes are often slower, and the implications of obsolescence are less critical.

• Ethical and regulatory standards: The medical field is heavily regulated by ethics committees, public health authorities, and regulators such as the Food and Drug Administration (FDA) or the European Medicines Agency (EMA) (Townend, 2017). If a study, drug, or treatment is withdrawn from the market or modified, this must be immediately reflected in the scientific literature to avoid misinterpretation. This creates additional pressure for medical publications to remain up to date and always reflect the best available evidence.

This article is a contribution to the burgeoning field of retraction studies. When the findings of a publication are intentionally or unintentionally compromised, and a correction is not sufficient to ensure their reliability, the publication may be subject to retraction. To date, Retraction Watch’s reference database contains almost 40,000 retractions.

We investigate retractions through the lens of gender: Does the author’s gender—whether the publication is written by a woman, a man, or a collaboration of both—affect the frequency or reasons for retractions?

Based on a cross-analysis of the Retraction Watch database and data from OpenAlex, this study shows first, that regardless of gender, single-author publications have a significantly lower level of retraction than multiple-author publications. This first finding has as much to do with the author’s ability to control their own publication as with the type of publication itself, which is generally more theoretical than empirical.

Second, when we switch from single-author to multiple-author publications, our findings reveal that mixed-gender teams have a higher probability of retraction compared to single-gender teams, suggesting that gender dynamics within research collaborations might introduce complexities that can affect oversight and responsibility.

Third, this study confirms, on the scale of all retractions and a wide variety of disciplines, the results obtained by the very few studies focusing on the relationship between gender and retraction mainly in biomedical sciences: Male-led publications are disproportionately retracted for serious ethical breaches, such as data falsification and plagiarism, while female-led publications are more frequently retracted for procedural issues, including methodological updates and textual errors. When women lead research teams, the reasons associated with retractions are not as severe as when these teams are led by men. This observed difference in the reasons for retractions can be attributed to gender differences in the practice of scientific work, but also to differences in the perception and evaluation of its outcomes.

These findings allow the articulation of two key institutional priorities: gender diversity and scientific integrity. It is well known that scientific institutions are implementing proactive policies to ensure greater representation of women in fields such as physics, mathematics, engineering, and computer science—disciplines where the presence of women is particularly limited. Similarly, scientific institutions are actively promoting best practices to ensure the reliability of the knowledge they produce.

Our article argues that these two priorities can be mutually reinforcing, using the presence of women to enhance the integrity of research and vice versa. As evidenced by publications resulting from mixed-gender collaborations, the severity of retractions tends to be lower when women are listed as first authors than when men occupy this role. Thus, it is not gender diversity alone that matters, but the positions that women hold within collaborations.

An institutional policy focused on reducing serious breaches of scientific integrity will view promoting women to positions of leadership as a way to foster good research practices. Conversely, a policy aimed at increasing women’s presence in science will see promoting best practices as an incentive for elevating women to leadership roles.

Promoting equity in science therefore goes beyond ensuring social justice or expanding the research questions—it is also about increasing the accuracy of the scientific record.

While rigorous checks and validations were conducted to ensure the quality of the OpenAlex data set, some limitations remain. It is possible that the data set contains inconsistencies or inaccuracies in author metadata, such as incomplete records, incorrect attributions, or ambiguous names, which could impact the gender identification process for certain publications. Such issues might lead to minor misclassifications or omissions, particularly in cases where the metadata was not detailed enough to assign a gender reliably.

However, these potential limitations are mitigated by the large size of the sample and the overall robustness of the data. The gender distribution by discipline, as shown in Figure 2, aligns well with established patterns and findings in the existing literature. This correspondence provides strong evidence that, despite the potential for some inaccuracies at the individual level, the data set as a whole captures broader trends effectively. As a result, we believe that the overall robustness of our data set allows for meaningful and reliable conclusions about gender composition and its role in retraction patterns.

Data available from The Center for Scientific Integrity, the parent nonprofit organization of Retraction Watch, is subject to a standard data use agreement. This paper is a significantly revised version of the preprint by Maddi, Monneau et al. (2023), incorporating extensive constructive feedback from Quantitative Science Studies reviewers.

Abdelghani Maddi: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Validation, Visualization, Writing—original draft, Writing—review & editing. Emmanuel Monneau: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Writing—original draft, Writing—review & editing. Catherine Guaspare-Cartron: Conceptualization, Formal analysis, Investigation, Methodology, Writing—original draft, Writing—review & editing. Floriana Gargiulo: Conceptualization, Data curation, Investigation, Methodology, Visualization, Writing—review & editing. Michel Dubois: Conceptualization, Formal analysis, Methodology, Project administration, Supervision, Writing—original draft, Writing—review & editing.

The authors have no competing interests.

This work was supported by a grant overseen by the French National Research Agency (ANR). Grant number: ANR-20-CE26-0008. Website: https://anr.fr/Projet-ANR-20-CE26-0008.

We used data from OpenAlex and Retraction Watch, both freely accessible via their respective websites (OpenAlex: https://openalex.org via API and Retraction Watch: https://retractionwatch.com).