Deciphering the role and quantifying the amount of work provided by different co-authors of a particular paper has been a recurrent problem for the scientific community [1,2,3]. The position in the list of authors is commonly used to infer co-authors’ contribution and a number of systems have been proposed on this basis. They range from simple calculations based on the rank of the authors such as harmonic authorship credit, fractional authorship credit and inflated authorship [4] to more complex credits (e.g. [5]), some even taking into account the controversial journal impact factor [2]. However, these metrics are essentially ‘one fits all’ approaches that assume the contribution of each author based on their position in the author list and attributes subjective and unfounded values to these positions. As such, they do not attempt to represent and quantify ‘true’ contribution. Despite the growing interest in resolving the issue of authorship contributions in scientific disciplines [3, 4, 6], no standard ranking system has been widely recognised or adopted by scientific journals. With this lack of consensus, some journals have implemented a compulsory or recommended section dedicated to reporting authors’ contribution.
A review of the top 150 ecology journals referenced in ISI Web Of Knowledge revealed that 13.3% of them require information on author contributions (Additional file 1). Authors are usually asked to briefly describe which task was conducted by which co-author. Although this information is valuable, it does not provide an objective, straightforward and universal measure of author contribution. For example, ‘data collection’ for a review article may in some cases simply involve searching a database by filtering papers using specific key words, while it may be a very time-consuming task in field ecology, and a highly technical task in computational ecology. So ‘data collection’ can mean very different things depending on the field of study or the type of paper. In addition, individual tasks are often conducted by multiple authors but there is no way of knowing whether and to what extent one author has contributed more to them. Although some authorship contribution systems propose graded contributions for each task (e.g. lead, equal, supporting role in the CRedIT system [7], or the three-tier criterion proposed by the International Committee of Medical Journal Editor [8]), the lack of a quantitative value means these systems lack accuracy and produce qualitative data that is challenging to analyse or compare. The second common limitation is the complexity of the proposed metrics, which deters authors from providing data in the first place. A third major issue is the lack of fairness where often the lead or corresponding author can unilaterally decide on the order of the co-authors and the description of their contribution.
To address these shortcomings, we propose an easy to apply, universally comparable and fair tool to measure and report author contribution.
A simple and accurate measure: percentage contributions
Percentages are straightforward and can be universally applied independent of research field, the number of co-authors or the nature of the paper (e.g. experimental, review, perspective, etc.). Because the authors of a paper are the best placed to make a judgement call about the value of each contribution, it is essential that percentage contributions are determined by authors rather than by a model based solely on the authors’ rank. Although disagreement may occur between co-authors, clarifying contribution among co-authors in the early stages of the research process is likely to ease potential tension [9] and in some cases prompt ‘real collaboration’. We propose that co-authors discuss and agree on their respective contributions prior to submitting their manuscript and these figures be provided by the corresponding author at the submission stage. By confirming their authorship, all co- authors confirm their agreement with their contributions and that of all other authors. This ensures that every published paper displays percentage contributions that have been discussed and agreed upon by every co-author.
A number of guidelines and best practices have been proposed for authors’ contributions (e.g. [10]). A possible starting point is to divide 100% by the number of authors and then estimate whether and to what extent each author provided more or less work than the others. The use of author-provided percentages has been proposed before to reflect the contribution of co-authors accurately (e.g. [2]) but with limited guidance about how to implement it. Verhagen et al. [11] developed the Quantitative Uniform Authorship Declaration (QUAD) approach, where each author is attributed percentage contributions in four categories: Conception and design, data collection, data analysis and conclusion and manuscript preparation. More recently, a very similar approach was proposed, based on scores rather than percentages with the more specific aim of deciding which contributor deserves authorship and which does not [12]. Clement [13] also suggests the use of four categories, albeit slightly different ones (ideas, work, writing and stewardship). However, an overly complicated metric may deter authors from applying it, and the criteria used in calculations must be consistent or have comparable importance across research fields, which may not be applicable to every type of article. In addition, many authors suggest that contributions should be restricted to an arbitrary threshold, for example 50% of the average contribution [13], 10% of the total work [11] or a threshold chosen by the authors [12]. Such limitation is likely to introduce major inconsistencies between papers, journals and fields of research, thereby preventing comparison. In addition, these thresholds limit the number of co-authors, which may affect interdisciplinary research and act as incentives to leave out minor contributors, potentially increasing ghost authorship (i.e. the omission of collaborators who did contribute to the work).
We propose that the contribution of each co-author be summarised in one number which must be more than 0% and less than 100% in multiple-authored papers. This provides a metric that is simpler for authors to determine and for the readers to grasp. In addition, this single metric imposes no upper limit on the number of authors. The percentage contribution should be displayed on the published paper either as raw numbers or as a figure (Fig. 1).
A universally comparable metric: percentage-based author contribution index (ACI)
An outstanding limitation of percentage contributions is that they are difficult to compare across different papers because with more co-authors, it is mathematically difficult to obtain high percentages. As a consequence, author contributions cannot be directly compared between articles with unequal numbers of authors. To allow such comparisons, we propose a universal metric that takes into account the number of co-authors: the Author Contribution Index (ACI), calculated from the percentage contribution as per Eq. (1).
$$ \kern0.5em \mathrm{ACI}(i)= Ci\times \frac{n-1}{1- Ci} $$
(1)
where for author i,
Ci = contribution of author i in percentage (must be > 0 and < 1)
n = total number of authors including i (must be > 1).
ACI(i) reflects the contribution of author i as compared to the average contribution of all other authors. It is superior to one when the contribution of author i is larger than the average contribution of all other authors and inferior to one when the contribution of author i is less than the average contribution of all other authors. For example, on a paper written by three authors, where author i contributed 60% of the paper, ACI(i) = 3, meaning that author i contributed three times more than what the other authors contributed on average. Another useful metric is log10(ACI), which is positive when the author’s contribution is larger than the average contribution of all other authors and negative when the author’s contribution is less than the average. This metric is particularly useful to normalise data for further comparison and statistical analyses.
The graph in Fig. 2 displays the universe of all possible ACIs for papers written by up to 200 co-authors. The contribution profile of a particular author can be displayed in the universe of possible ACIs by adding dots, each representing one paper from the author being analysed. From these data, author profiles may appear according to a variety of criteria such as time, author’s seniority, area of research and type of institution where the author works, among others. Based on Eq. (1), it is also possible to calculate average ACI for an individual author or to plot the ACI frequency distribution of an individual author based on all or specific parts of his publications.
One notable feature of ACI is that it increases with the proportion of work produced but also with the number of ‘minor’ co-authors (Fig. 2). By giving more weight to main contributors of papers with many co-authors, ACI recognises the skills required and work involved in leading large collaborative projects. Figure 3 provides examples of how ACIs could be displayed in a paper.
A fair tool: assisting job seekers, recruiters and performance-based evaluations
The scientific community seems to have reached the consensus that journal impact factors are not an accurate measure of the value of a particular article or the value of its author(s) [14]. One of the main reasons is that a very highly ranked journal may publish few articles that are heavily cited, but it may also publish a number of papers that have very little impact. In recent years, article-based impact has been preferred to journal impact factor. For example, the Hirsch index (h-index), which is based on the number of citations of one’s papers, is now widely used to gauge the output of a scientist. However, the h-index can also be manipulated [15] and it does not provide a measure of the amount of work produced by each co-author, which means guest or honorary authorship (i.e. inclusion of authors who did not contribute to the work) cannot be accounted for. This issue of guest authorship has been denounced widely in medical and clinical science [16], but other research fields are not immune to the problem (e.g. in ecology [17], environmental science [18], geography [19], geology [20], etc.). Since genuine authors suffer no cost when they add co-authors, papers tend to have more and more co-authors [21, 22]. With percentage contributions, the amount of work invested in a paper is a finite value (100%). Therefore, when more authors are added as a ‘gift’, they all need to be attributed a percentage of the work. In this zero-sum game, either it will be visible that guest authors have contributed an extremely small proportion of the work—and should receive very little recognition—or the genuine authors will have to give away large chunks of their well-deserved credit.
To ensure fair values are reported, we propose that co-authors discuss and agree on their respective contributions prior to submitting their manuscript and these figures be provided by the corresponding author at the submission stage. By confirming authorship, all co-authors confirm their agreement with their contributions and that of all other authors. This ensures that every published paper displays percentage contributions that have been discussed and agreed upon by every co-author.
The sentence-based descriptions of authors’ contribution that are used by some journals provide an indication of tasks performed by each author. Maintaining and generalising this practice as a detailed record of the role of each author is essential to increasing transparency. However, obtaining a clear idea of the amount of work a scientist is actually providing is difficult if one needs to read through all the authors’ contribution sections and weigh in the topic, the type of paper, the number of co-authors, etc. The proposed index has the potential to complement descriptive authorship information sections with a quantitative measure, which is much easier to analyse, summarise and compare across a large number of publications. For example, ACI can help in sifting through the numerous papers published by one scientist to select only those where this particular author has made a major contribution. This short list of papers could then be analysed in more details, for example using the written author contribution sections.
ACI can provide valuable information for performance-based evaluation processes and could be implemented in existing reporting systems. This includes internal evaluation for career advancement, as well as research productivity evaluation for funding purposes and national-scale ranking schemes (such as the Performance-based Research Fund (PBRF) system in New Zealand or the Research Excellence Framework (REF) in the UK). It is also in the advantage of the candidate to be able to demonstrate his/her actual contribution to a potential recruiter who may ask ‘What have you done on all these papers listed on your CV?’ One could answer such a question by analysing the distribution of a scientist’s ACI and its evolution through time or by calculating and comparing his/her average ACIs in experimental, review and perspective papers. ACI could also be used as an additional metric in network-based collaboration analyses (e.g. [23]) or to further inform composite citation indicators (e.g. [24]).
A first look: testing ACI
Aims
Here, we provide an analysis of ACI calculated from work published in the past 3 years (January 2014–December 2016) based on contribution percentages provided by scientists who volunteered to respond to an online survey (Additional file 2). This survey aims at (1) demonstrating that authors can provide percentage contribution for their work and are willing to do so, (2) demonstrating that there is an interest from the scientific community to provide more transparency in author contribution, (3) exemplifying basic calculations that can be easily performed using ACI values and (4) demonstrating that ACI can be used to understand the scientific production of scientists in relation to their career advancement. Because the contribution percentages were provided after publication and without discussion among co-authors, these values may not be as accurate as if they had been agreed upon by all co-authors prior to publication [10]. Hence, the aim of this exercise was not to produce a highly accurate dataset, and therefore, the following analysis should be regarded as illustrative.