Beyond authorship: Analyzing disciplinary patterns of contribution statements using the CRediT taxonomy

This is a preprint. It has not undergone peer review. The final published version may differ. 1 Beyond authorship: Analyzing disciplinary patterns of contribution statements using the CRediT taxonomy Elvira González-Salmón1, Victoria Di Césare1*, Aoxia Xiao2 and Nicolas Robinson-Garcia1 1Unit for Computational Humanities and Social Sciences (U-CHASS), University of Granada (Spain) 2School of Information Management, Wuhan University (China) *Corresponding author: [email protected] Abstract In this research article we present the first cross-disciplinary descriptive analysis on the use of contribution statements. Our main objective is to obtain further insight on contributions by a variety of fields (Multidisciplinary, Health, Life, Physical and Social Sciences) from the largest dataset used up to now. We examine more than 700,000 articles published between 2017 and 2024 in Elsevier and PLOS journals, in combination with bibliometric data extracted from the Scopus database. The descriptive analysis of the dataset focuses on the overall coverage of the merged data, the distribution of authorship and disciplines at paper level, and the interactions between contribution statements, author order and disciplines. Our two main findings indicate that, on the one hand, looking at contributions and authorship order can enrich the way we understand science as a social endeavor. On the other hand, delving deeper into contributorship differences by field is key. We underscore the value of the CRediT taxonomy in unveiling nuanced research dynamics and offering a more equitable framework for evaluation. Keywords: CRediT taxonomy; Authorship; Contributorship; Division of labor Introduction Team science is on the rise (Wuchty et al., 2007), challenging previous conceptions of what accounts for science and authorship (Birnholtz, 2006). This may require new norms that are prepared to deal with the new structure of science (Jabbehdari & Walsh, 2017) and has already led many to suggest replacing the concept of author with that of contributor, which acknowledges the distributed and collaborative nature of science as it is conducted in the 21st Century. Since their introduction in biomedical journals by the end of the 1990s (Rennie et al., 1997; Rennie et al., 2000), contribution This is a preprint. It has not undergone peer review. The final published version may differ. 2 statements are becoming more widespread in academia, especially thanks to the launch of the Contributor Roles Taxonomy (CRediT) (Brand et al., 2015). This taxonomy was developed to homogenize contribution statements across publications and facilitate discussions about authorship that help avoid author disputes (Allen et al., 2019). Since its launch, the CRediT taxonomy is now used by more than 40 different publishers across a wide range of disciplines (https://credit.niso.org/). Their expansion not only has the potential of overcoming the limitations of using authorship as a “credit” in research assessment but can help us understand how science is done, how researchers distribute tasks, how this influences author order, and how it relates to their prestige as reflected in the author byline of publications (Haeussler & Sauermann, 2013). Contributorship has been studied for some years now. Cronin et al. (2003) traced them even before they appeared in a special section within research articles by looking into the acknowledgements for Psychology and Philosophy throughout the 20th Century. Drawing on a multidisciplinary dataset of more than 80,000 documents, Larivière et al. (2016) examined the relationship between division of labor, contribution types and authors’ seniority to provide evidence on the existence of conceptual contributions made by senior researchers and technical tasks performed by younger scholars. Later, Larivière et al. (2021) updated their analysis using the CRediT taxonomy for a set of more than 30,000 PLOS papers and studied the distribution of contributions across teams. Other relevant works have applied contributions to look into credit allocation (Ding et al., 2021; Sauermann & Haeussler, 2017), credit bias (Matheson, 2022), scientific trajectories (Robinson-Garcia et al., 2020), and gender differences (Macaluso et al., 2016; Sugimoto & Larivière, 2023). However, much remains unknown since most analyses focus on specific disciplines and publications, mainly Biomedical Sciences and PLOS journals. In addition, from a bibliometric point of view, the possibilities of this still novel taxonomy are not fully explored. More than one decade since the birth of CRediT, we now have enough data to start getting a glimpse of its adoption, patterns and prospects. Here we present the first cross-disciplinary descriptive analysis on the use of contribution statements. The aim of this descriptive paper is to obtain further insight on contributions by a variety of fields, covering Health, Life and Physical Sciences as well as Multidisciplinary and Social Sciences, from the largest dataset used up to now, to the best of our knowledge. We analyze a set of over half a million research articles belonging to Elsevier and PLOS journals from all fields of science which we combine with bibliometric data extracted from the Scopus database. We look into disciplinary differences on the use of contributions, their overall coverage and their relationship with author position in papers. We conclude by discussing the potential of this data to open new venues of research on career trajectories, scientific impact and recognition, and the social organization of the sciences. This is a preprint. It has not undergone peer review. The final published version may differ. 3 Data & Methods In this study we examine a total 714,732 journal articles published between 2017 and 2024. The data was facilitated by the ICSR Lab from Elsevier as well as by PLOS. Elsevier was one of the first publishers adopting the CRediT taxonomy within their journal portfolio (Elsevier, 2024). Since then, they provide the option to include such information in their Editorial Manager and are working on offering it as well through the ScienceDirect journal platform (Genova, 2023). The ICSR Lab provided access to a total of 633,443 unique articles from 1,951 journals, authored by a list of 1,854,552 unique author profiles ranging from 2017 to 2024. We conducted some manual quality checks to ensure the reliability of the dataset which are available in the Appendix (see Table A1). PLOS data was also facilitated by the publisher through a data use agreement. PLOS has stand out as one of the main drivers on the expansion of contribution statements, by liberating their bibliographic data and allowing the scientometric community to explore it to better understand team dynamics. Examples of such efforts are the studies conducted by Larivière et al. (2016, 2021), Sauermann and Haeussler (2017), Macaluso et al. (2016) and Robinson-Garcia et al. (2020) to name just a few. In this case they provided contribution data from their journal portfolio related to 97,819 publications for the 2018-2023 period, from which 83.2% belonged to PLOS One. Along with the author contribution statements assigned to every author, this dataset includes each article’s Digital Object Identifier (DOI). After a manual validation check (See Appendix, Table A2), the dataset was merged into Scopus data via DOI and authors’ given name and surname. A total of 81,289 records were correctly matched, corresponding to 415,014 disambiguated authors (based on Scopus Author Profile). Our final dataset resulting from the merging of the ScienceDirect data and the PLOS data contains a total of 1,965 journals and 714,732 bibliographic records, authored by a total of 2,182,041 unique Scopus Author Profiles and published between 2017 and 2024. We have a minimum number of authors per paper of 1, and a maximum of 271. The full Python code developed to merge, compute and visualize the data is freely available in Di Césare and Xiao (2024) and the supplementary material can be accessed through González-Salmón et al. (2024). Next, we conduct a descriptive analysis of the dataset focusing on the overall coverage of the merged data, the distribution of authorship and disciplines at paper level, and the interactions between contribution statements, author order and disciplines. Results General overview Figure 1 provides an overview of the magnitude of the dataset analyzed in comparison with the overall size of Scopus. The first three years of the period as well as 2024 present low coverage, but the This is a preprint. It has not undergone peer review. The final published version may differ. 4 years 2020 to 2023 represent on average 6% of Scopus’ content, with 2022 reaching the highest coverage point at 7.2% (208,207 articles). Table S1 (Supplementary material) shows this comparison by discipline and field. The list of fields, disciplines and their corresponding acronyms is included in the Appendix (Table A3). Figure 1. Coverage of our merged dataset compared to Scopus by year. Both in absolute and relative terms, the most represented field in our dataset and compared to Scopus is Physical Sciences (503,145 articles, 5% of Scopus’ content). Within it, the disciplines Engineering (198,231), Materials Science (161,105) and Chemistry (129,971) stand out with the highest numbers of articles for the whole period. Multidisciplinary as a field is also prominent because it presents the highest percentages of shared articles between our dataset and Scopus throughout the years (15% on average) (Supplementary material, Table S1). In addition, Table S2 (Supplementary material) lists the 1,965 journals that form our merged dataset with the number of articles each one contains per year. The journal covering the largest number of publications from our dataset is PLOS One, with almost 10% of the total articles. Paper level analysis Figure 2 shows the distribution of authors per paper overall and by major field. To ensure readability, we include a cut-off threshold of up to 50 authors per paper. The mode for the whole set is of 4 authors per paper (113,991 papers). When looking at number of authors across fields, we observe narrow differences. For Social Sciences, the most common number of authors is 3 (16,150 papers), This is a preprint. It has not undergone peer review. The final published version may differ. 5 while for both Multidisciplinary (9,772 papers) and Physical Sciences (85,592 papers) it is 4. In Health Sciences and Life Sciences most papers are written by 5 authors (9,570 and 23,321 papers respectively). Overall Health Sciences Life Sciences Multidisciplinary Physical Sciences Social Sciences Figure 2. Distribution of number of authors (<= 50) by number of articles and field, log scale. Similarly, in Figure 3 we examine the distribution on the number of contribution types per paper overall and by field. All fields except for Multidisciplinary follow a similar pattern, in which the number of papers decreases after reaching an average peak of 9 contribution types per paper. In Social Sciences the maximum number of contribution types per paper is 8, whereas Physical Sciences has 9, Health Sciences and Life Sciences have both 10, and Multidisciplinary 11 contribution types. Thus, Social Sciences is the field with the fewest contribution types, while Multidisciplinary presents almost 40% more contribution types. Overall Health Sciences Life Sciences Multidisciplinary Physical Sciences Social Sciences Figure 3. Distribution of number of CRediT contribution types per paper, overall and by field. The average proportion of contribution types by discipline can be seen in Figure 4. Some particular statements, like Conceptualization, Methodology, Writing – original draft and Writing – review & editing, are clearly present in all disciplines at a high rate. Others, such as Resources, Project administration and Funding acquisition, are conversely seldom present. In between, we find a few mixed scenarios where the same contribution, for instance Software, Data curation and Formal analysis, has a high proportion of usage in some disciplines but very low in others. This is a preprint. It has not undergone peer review. The final published version may differ. 6 Figure 4. Average proportion of CRediT contribution types by discipline. Author level analysis Figure 5 shows the average number of contribution types each author contributes with on average per field. In this way, we show how distributed are tasks by field. For example, in Physical Sciences authors contribute on average with one task in 13.84% of the articles, whereas in 30.40% of them they conduct an average of 2 different tasks. In Health, Life and Physical Sciences the most common scenario is that of 2 contributions per author (26.52%, 27.80% and 30.40% respectively), while in Multidisciplinary and Social Sciences each author performs 3 tasks in 20.30% and 23.20% of the papers. If we look at higher numbers of contributions per author, we find that Health, Life and Physical Sciences are less represented in those groups (for instance, in less than 3% of their articles authors contribute with up to 7 different tasks), while Multidisciplinary and Social Sciences have a slightly more significant representation there (6.28% and 4.59%). Therefore, we observe a higher distribution of tasks in fields such as Health, Life and Physical Sciences. On the opposite side we find the Multidisciplinary field and Social Sciences, where authors are involved on a higher number of tasks on average. This is a preprint. It has not undergone peer review. The final published version may differ. 7 Figure 5. Average number of CRediT contribution types by author involved per article and field. Co-occurrence of contributions by paper Here we investigate the level of co-occurrence between contribution types, that is, how common it is for a contribution to be conducted by the same author in the same publication (Figure 6). The most common contribution co-occurrences take place between Conceptualization and Methodology (0.40), Writing - review & editing (0.35), and Writing - original draft (0.34). Furthermore, Writing - original draft, also tends to co-occur with Methodology (0.34) and Formal analysis (0.31). The contributions that appear together the least are mostly related to Software and other statements (e.g. Funding acquisition, Supervision and Project administration). The pairs formed by Funding acquisition and Data curation, Funding acquisition and Visualization, and Supervision and Data curation are also not usual contributions done by the same author within a publication (0.07 across all of them). There seems to be a higher co-occurrence of conceptual contributions whereby those who are involved in management related contributions (e.g. Resources and Project administration) do not typically perform technical tasks (e.g. Data curation and Software). When looking into specific fields, we observe that Methodology and Conceptualization is the most common combination in the Social (0.52), Health (0.41) and Physical Sciences (0.39), while Writing - review & editing together with Conceptualization co-occurs the most in Multidisciplinary (0.46) and Life Sciences (0.37). In general, there is a higher co-occurrence of contributions in the Social Sciences and Multidisciplinary fields than in the rest. This finding is consistent with the results shown in Figure 5, which indicate that authors in these fields are more likely to contribute to multiple roles simultaneously. This is a preprint. It has not undergone peer review. The final published version may differ. 8 Overall Health Sciences Life Sciences Multidisciplinary Physical Sciences Social Sciences Figure 6. Co-occurrence of CRediT contribution types by field. Relation between author order and contribution type Finally, we analyze the relation between contributions and author positions. We do so by looking at the percentage of first, middle and last authors that perform each contribution overall and by field (Figure 7). First authors perform a higher number of contributions (37% on average) than last authors (27%) or middle authors (19%). The most common contribution among first authors is Writing - original draft (83.5%), followed by Methodology (67.2%), Conceptualization (63.3%) and Investigation (58.1%). Conversely, the contributions in which first authors participate the least are Resources (9%), Funding acquisition (10.8%), Supervision (10.9%) and Project administration (11.5%). In the case of last authors, their most common contributions are Writing - review & editing (66.3%) and Supervision (51%), while their least common contributions are Software (6.9%) and Visualization This is a preprint. It has not undergone peer review. The final published version may differ. 9 (10.7%). As for middle authors, they participate the most in Writing - review & editing (45%) and Investigation (32.6%), but contribute very little to any of the rest (16% on average). The results by field display a similar pattern where first authors always conduct a higher number of contributions. In all cases their main contribution is Writing - original draft, followed by Conceptualization (all fields but Multidisciplinary), Methodology (in Health, Social Sciences and Multidisciplinary) and Investigation (only in Life and Physical Sciences). The least common tasks among first authors are the most common ones among last authors, such as Supervision (across all fields) and Funding acquisition (in Health, Life Sciences and Multidisciplinary). Still, last authors particularly stand out in Writing - review & editing and Conceptualization. Lastly, middle authors’ main contributions are Writing - review & editing (all fields but Multidisciplinary), Conceptualization (even with first authors in Health Sciences), Investigation and Methodology (especially in Multidisciplinary and Social Sciences), but they almost do not participate in Software, Visualization and Funding acquisition (with the exception of Physical Sciences in the latter). Figure 7. Relation between CRediT contributions and author positions by field. Discussion This paper presents the first cross-disciplinary analysis of contribution statements performed. The novelty of this research lies in the unique dataset we analyze, which is the largest used so far to compare results at the field, discipline and author levels. Furthermore, we include a large sample of First Middle Last First Middle Last First Middle Last 714671 2590052 701041 73723 364654 72821 164330 713629 162339 Conceptualization 63.3 21.8 48.0 68.3 23.4 62.5 61.9 60.5 42.7 Data curation 38.9 19.8 12.2 43.8 24.5 17.1 41.5 19.7 13.1 Formal analysis 49.0 21.0 16.5 57.9 21.0 21.9 56.1 21.8 18.6 Funding acquisition 10.8 10.2 32.6 11.9 8.0 33.0 11.5 9.9 42.7 Investigation 58.1 32.6 19.2 54.2 29.0 22.8 62.2 29.1 19.8 Methodology 67.2 27.8 30.1 65.3 29.0 40.4 66.4 29.4 33.8 Project administration 11.5 8.4 25.8 17.2 8.4 29.2 12.2 7.6 32.7 Resources 9.0 14.7 20.9 10.9 15.9 21.3 8.7 15.8 23.8 Software 25.4 9.2 6.9 18.2 7.2 5.7 19.8 8.1 12.1 Supervision 10.9 19.2 51.0 12.0 16.0 59.1 9.4 16.4 61.2 Validation 24.7 15.6 16.7 20.9 13.2 17.1 15.8 13.2 15.4 Visualization 29.4 10.6 10.7 32.0 9.9 12.9 41.9 10.2 12.1 Writing – original draft 83.5 14.1 18.8 86.4 15.6 27.0 85.1 13.4 24.2 Writing – review & editing 39.9 45.0 66.3 47.9 54.5 75.0 41.9 43.6 72.0 Life Sciences % of authors by CRediT Author number Author position Field Overall Health Sciences First Middle Last First Middle Last First Middle Last 69564 305681 68416 503129 1654716 493423 69912 160627 67060 Conceptualization 72.0 30.5 70.7 61.2 40.6 40.6 77.5 46.4 46.4 Data curation 61.9 31.1 26.1 34.7 17.0 9.3 38.0 20.4 15.0 Formal analysis 74.5 27.6 33.1 42.7 19.3 12.7 54.1 21.7 19.4 Funding acquisition 20.2 10.5 40.7 36.6 28.1 30.3 15.5 11.9 23.0 Investigation 66.0 36.3 34.1 56.8 30.9 16.3 47.1 32.1 20.2 Methodology 75.5 37.3 49.9 54.5 24.9 24.9 76.0 32.1 35.6 Project administration 32.2 12.7 36.9 7.3 12.6 23.0 15.5 9.7 17.3 Resources 18.0 16.0 26.8 36.8 14.0 19.7 10.7 12.2 15.7 Software 9.0 8.0 15.7 36.8 10.2 21.9 31.0 11.1 9.3 Supervision 20.7 21.7 71.0 14.2 20.0 54.1 14.1 22.0 34.2 Validation 16.3 16.3 25.5 14.4 16.3 16.3 16.1 16.1 17.8 Visualization 37.1 9.8 15.7 27.5 10.8 9.3 30.9 12.1 11.6 Writing – original draft 91.5 18.2 37.4 81.6 12.2 13.6 83.6 22.7 23.6 Writing – review & editing 64.6 0.1 82.8 34.6 34.6 61.7 48.9 56.5 67.9 Author number % of authors by CRediT Social Sciences Physical Sciences Multidisciplinary Field Author position This is a preprint. It has not undergone peer review. The final published version may differ. 16 Social Sciences Arts and Humanities ARTS Business Management and Accounting BUSI Decision Sciences DECI Economics, Econometrics and Finance ECON Psychology PSYC Social Sciences SOCI Multidisciplinary MULT